Hedge-Fund AI Podcast and Video Practitioner Signals

Related research: GMO, Acadian, Arrowstreet, and selected managers · Hedge-fund research-machine leakage · Finance and quant audio/video roadmap · External Media Research Room

New source note: title-blind finance-media records — Goldman QIS, former Citadel AI researcher, and Versor

New title-blind capture: Optica / Adarsh Nair — Beyond the Grind

New first-party manager route: Binomial Technologies — public AI research stack and Sentry

New title-blind practitioner route: BedRock Partners E19 — Bill Sun on AI application in investment

Additional capture note: Versor partners — Odds on Open

Conference/vendor addendum: CQF, QES, and Quantmate agentic-finance media surfaces

LSEG archive addendum: Hedge Fund Huddle archive expansion — Versor/StarMine, LLM adopters, and Epistemic AI

LSEG title-blind addendum: Alternative data and low-latency/data-sourcing archive follow-up

LSEG AI/performance addendum: Title-blind AI, sentiment, and performance-data follow-up

LSEG fixed-income/regional addendum: Fixed-income automation and Middle East regional coverage

September 2 title-blind addendum: Man Group CIO and Blushing Quants #13 — research-scale and quant-pipeline routes

September 2 transcript-retention addendum: Blushing Quants AI and quant transcript-gap recovery — five public enclosures now have durable private TXT/SRT artifacts.

September 2 HFF recovery addendum: Hedge Fund Huddle transcript-gap recovery — ten additional public enclosures now have durable private TXT/SRT artifacts, including a Man Group CIO episode, a finance-LLM adoption panel, an Epistemic AI discussion, and title-blind operating-context episodes.

September 2 coverage-frontier addendum: Firm media, personnel, regional, and research-lineage routes

September 2 emerging-firm and India-quant addendum: Emerging-firm, regional, and title-blind audio routes

September 2 allocator-research addendum: Zenith Investment Researcher’s AI in funds management episode

New QuantSpeak capture: Renee Yao / Neo Ivy explainable-ML interview

QuantSpeak methodology captures: RAM AI, Wright Research, and J.P. Morgan interviews

QuantSpeak vendor/data-modality capture: Grant Fuller / Irithmics on vicarious risk

QuantSpeak early AI/ML archive: NVIDIA, volatility-surface research, Acadia, and Paul Wilmott

QuantSpeak manager/risk archive: Pine Tree, SCT Capital, Vola Dynamics, and ITO 33

QuantSpeak archive-completion screen: 14 remaining episodes, five promoted lanes, and reviewed archive coverage

Executive answer

The expanded pass added firm-controlled YouTube, publisher transcript, podcast-index, and guest-name discovery to the existing RSS workflow. It surfaced direct practitioner evidence for HRT, Jane Street, Balyasny, Numerai, Versor, Point72, Two Sigma, CFM, and Voleon, plus regional comparison evidence from RQI Investors in Australia and TD Asset Management in Canada, and adjacent evidence from QIS and market-making sources.

The August 17 follow-up adds a separate set of public signals: Kai Wu’s GMO-to-Sparkline lineage and NLP/alternative-data factor work; Stoic Point’s screening, research, and monitoring workflow; Epoch’s deterministic research interface; OneEye’s options-specific ML vocabulary; and methodology episodes that separate predictive ML from generative text systems. The August 18 title-blind expansion adds transcript-backed records for OneEye founder Eren Biri, GenAI/quant practitioner Denis Lukyanov, and former Citadel practitioner Jerome Busca. The follow-up note keeps these items separate from the earlier firm matrix and marks queue-only leads explicitly.

The same title-blind pass also recovered a Chicago Global episode with Ben Charoenwong, whose public account connects mid-frequency, ex-US investing and alternative data with foundational tabular-model monitoring, LLM-assisted data structuring, feature engineering, and deterministic signal controls. A direct publisher MP3 and local timestamped ASR pass now support navigation for this episode and the related Blushing Quants entries. This is a named practitioner methodology disclosure, not evidence of a complete Chicago Global model inventory or performance result.

The latest title-blind video pass added two distinct records. A 2022 Mike Chen interview captures his then Robeco role and a historical discussion of alternative-data families, web collection, NLP/ML processing, hypothesis-first data selection, and overfitting risks (03:11–08:20; 12:15–16:51; 21:52–24:11). The publisher labels the transcript AI-generated, so this remains historical role-and-method evidence rather than a current Robeco system claim. A separate Silvercrest Asset Management recording, published August 21, 2026, names Chad Kusserow and Jeff Nevins; Silvercrest’s first-party announcement confirms their roles. Automatic captions discuss algorithmic trading, factor flips, AI-related repricing, social-media information, and collaborative data-point validation (01:31–06:40; 12:10–15:20). The Silvercrest route is an adjacent institutional-manager comparator and does not disclose a model, training corpus, vendor, permission map, execution rule, or AI-attributed result. Both records and their boundaries are documented in the capture note.

Wolfe QES: agenda-level AI, alternative-data, and quant-research signals

The Wolfe EventSmart archive and venue index expose more than event names. The supplied 2025 AI in Finance landing page explicitly frames the event around DeepSeek R1, GPT-4, and Claude, and names AllianceBernstein, Goldman Sachs, T. Rowe Price, and Cubist Systematic as participating institutions. The linked agenda then names research routes spanning earnings-call Q&A, LLM-assisted core-earnings measurement, dynamic knowledge graphs, agentic LLMs, AI-agent collaboration, GenAI in conference calls, continuous-audit cross-verification, and open multimodal financial LLMs. The organizer biography for Yin Luo adds a dated public personnel route: Wolfe identifies him as its quantitative research, economics, and portfolio-strategy leader and describes prior global quantitative-strategy leadership at Deutsche Bank. None of these pages identifies the chief-AI-officer panelists or discloses a named institution’s systems, data permissions, production use, or results.

The 2023 Global Quantitative and Macro agenda supplies a useful earlier vocabulary: AI-powered trading and algorithmic collusion, human-plus-machine stock analysis, deep-neural-network macro dynamics, machine-factor models, transcript-based corporate-risk extraction with generative AI, and sentiment risk premia. The 2025 European agenda adds conditional time-series GANs, graph-clustering statistical arbitrage, agentic and multimodal thematic investing, random-forest macro forecasts, and LLM sentiment allocation with proximal-policy optimization. These are dated conference topics and named speaker associations, not evidence of a fund’s production implementation or performance.

The missing 2024 bridge is now captured through the 8th Global Quantitative and Macro agenda. It names a “Triple Sentiments” session combining analyst ratings, textual NLP sentiment, and speech emotions; LLM macro forecasting; ML for market microstructure; uncertainty-aware prediction; and alternative-data research on corporate R&D following competitors’ drug-development disclosures. That creates direct research routes into voice features, macro-LLM evaluation, market microstructure, and clinical-trial/regulatory-event prediction. The agenda is a public speaker/topic map, not evidence of a named fund’s live model, data rights, or portfolio authority.

The QES job-postings luncheon is a separate alternative-data disclosure: Wolfe says the presentation used RavenPack data covering more than 200 million job postings from more than 60,000 companies across 195 countries since 2007, with more than 10,000 skillset metrics. The page describes factors around hiring trends, job growth, geographic expansion, staffing demand, and technical skills. Those figures and descriptions are Wolfe’s public event claims; the page does not disclose licensing, point-in-time controls, feature construction, or a reproducible backtest.

CFM: internal technology-offsite agenda and named practitioner interview

CFM’s Tech Offsite 2026 page is currently locked, but its public Digitevent activity endpoint exposes a 40-plus-item agenda. The research-relevant entries include separate sessions for AI in research, GenAI for feature generation and data analysis, a future-of-coding panel, high-scale entity matching, security guiding AI transformation, a technology–research fireside chat with Alexios Beveratos, collective intelligence with Christopher Pratt, and “When anyone can write code” with Ian Howard. No replay URLs, speaker roster for the AI sessions, documents, model details, or production evidence were exposed. The capture note excludes operational credential content from the public record.

CFM’s official YouTube channel adds a title-blind historical media route. The 2019 Columbia alternative-data initiative video uses automatic captions to describe CFM’s seminars/workshops with Columbia’s Program for Economic Research and satellite-derived inventory/storage estimation as an example (00:28–00:43; 01:38–01:56). The 2026 Philip Seager at Milken recording has no public English captions in the checked route, so it remains metadata-only. These are historical partnership and event signals; they do not establish a current dataset license, model, collaboration, or investment result. The channel capture note records the checks.

The separate PyData Paris 2024 interview names CFM data scientists Lamine Souiki and Marine Michaut. Their public transcript describes Python in model development, execution, data exchange, and reliability work; classical ML and foundation models for time-series forecasting and NLP; scikit-learn, PyTorch, OpenAI, Mistral, Llama, NetworkX, and AWS managed services. This is named practitioner evidence and tool vocabulary, not a complete CFM model inventory or proof that every named tool is used in a live investment process. See the capture note.

The older Risk.net Quantcast interview with Jean-Philippe Bouchaud, now recovered from its SoundCloud track, adds a distinct historical layer. Bouchaud describes CFM’s scientific research lineage and names price, fundamental, weather, inventory, machine-learning, and natural-language-processing inputs as possible signal sources (03:04–04:02). He presents agent-based models as scenario generators for risk management and as a way to investigate unstable feedback loops, explicitly separating that simulation use from a direct trading-agent claim (09:59–22:47; 22:45–22:47). In the machine-learning discussion, he places ML across portfolio construction, model allocation, execution, and signal generation and mentions cross-validation as a control on parameter selection (31:30–35:15). He also describes adding a new nonlinear model alongside an existing model and treating incoming data vendors as a screening and resource problem (35:15–36:15; 36:15–39:50). This is dated practitioner evidence with a local audio/ASR layer; it does not establish CFM’s current model inventory, data rights, agent permissions, production authority, or performance. See the audio recovery note.

Jane Street / In Young Cho: ML research, productionization, and model constraints

Jane Street’s first-party Signals & Threads episode with In Young Cho was already represented by a publisher/audio capture, but its firm-controlled YouTube route lacked a promoted transcript sidecar. The local recovery now adds a second timestamped navigation layer. The episode’s publisher describes Cho as helping lead the research group’s machine-learning work. The discussion covers data generation and feature/response construction, the move from linear models toward deep learning as interactions grow, productionization from notebooks into reliable trading systems, and the tension between flexible exploratory tools and robust production systems (14:41–20:36). Cho and Ron Minsky also describe short iteration cycles, model/data constraints, and the need to distinguish research flexibility from production reliability (20:36–23:10; 47:10–50:49). These are first-party practitioner and process statements; they do not disclose a complete model registry, training corpus, data rights, live permissions, autonomous order authority, or investment performance. See the capture note.

The queue audit also recovered the first-party Signals & Threads episode with Will Wilson from its Jane Street YouTube route (recording). The March 17, 2026 episode explicitly describes Jane Street as both a customer and investor in Antithesis (00:03–00:31), and Minsky describes applying the system to Jane Street’s internally developed Aria distributed system alongside existing deterministic-simulation testing (53:39–54:10). The conversation also treats AI code generation as increasing the need for verification and distinguishes exploratory software from systems where correctness is critical (57:15–58:10). These are engineering-control and vendor-relationship signals; they do not disclose a trading-model use, agent permissions, model providers, or performance. The alternate caption recovery is recorded in the capture note.

The earlier AI Engineer talk by John Crepezzi was already present in the repository, but the queue audit confirmed that it had not been reflected with sufficient detail in the firm synthesis. Crepezzi identifies Jane Street’s AI Assistance team (00:17–00:50), describes workspace snapshots and build-status transitions as training data (06:08–09:23), and describes a Code Evaluation Service that checks generated diffs against compilation, typechecking, and tests (09:23–11:37). He also describes AIDE as a sidecar for context construction, prompt and model switching, editor integrations, telemetry, and A/B testing (12:39–15:50), then names retrieval, multi-agent workflows, and reasoning models as ongoing work (15:50–16:16). This is official event-media practitioner evidence about internal developer tooling and evaluation design; it does not establish trading-model use, model weights, data rights, permissions, or performance. See the capture note.

The queue also promoted Jane Street’s Doug Patti episode on state-machine replication, published April 20, 2022. The publisher identifies Patti as a Client-Facing Tech developer working on Concord and Aria; the recovered recording adds discussion of audit trails, deterministic testability, replay, access controls, and bounded automation in distributed systems (03:03–03:34; 23:38–28:09; 48:32–50:35; 65:09–68:37). This is useful engineering-control context adjacent to the later Antithesis relationship, but it does not disclose AI/GenAI use, trading-model permissions, or performance. See the capture note.

The same title-blind recovery added the November 2015 Software Engineering Daily interview with Yaron (Ron) Minsky, with a timestamped recording. Minsky corrects the host’s title at 00:51–01:02, separating his Head of Technology role from quantitative research leadership at that date. He ties automated trading to real-time multi-source data, timestamped historical data, bulk research, rapid iteration, and correctness controls (01:07–03:40). He later describes risk-tiered engineering, redundant stop-trading paths, and code review as knowledge transfer and a constraint on system readability (44:00–44:26; 49:04–50:58; 53:07–54:43). This is historical external engineering evidence, not a current AI or GenAI disclosure; it does not establish model identity, agent permissions, data rights, trading authority, or performance. See the capture note.

The same pass recovered the distinct Strange Loop 2018 “Data Driven UIs, Incrementally” session and its timestamped recording. Minsky describes Jane Street’s internally used, data-driven trading interfaces (00:11–00:56), demonstrates dynamic operations over roughly 100,000 rows (01:17–02:21), and develops Incremental and diff/patch abstractions for efficient change propagation (08:41–10:14; 20:49–26:18). The close frames UI design as an optimization problem and diff/patch as a bridge between functional data structures, incremental computations, network protocols, and browser APIs (33:39–35:46). This is dated conference engineering evidence, not a current AI, GenAI, model-permission, or performance disclosure. See the capture note.

Magnetar: compute-for-equity and AI-infrastructure financing

The title-blind pass recovered the December 9, 2024 Bloomberg Odd Lots episode with Magnetar’s Jim Prusko. Magnetar’s first-party recap identifies Prusko as a Partner and Senior Portfolio Manager and links the discussion to Magnetar’s AI-infrastructure financing activity. The episode describes Magnetar’s first institutional investment in CoreWeave in 2021 and a financing thesis in which contracted compute can be paired with equity investment (approximately 04:10–05:11; 19:22–20:41; 29:49–30:44). It also discusses GPU-backed lending, energy and power requirements, and target company categories such as custom or small models, robotics, autonomous driving, weather models, and application-layer models (09:30–10:24; 14:40–15:10; 21:16–22:53).

Magnetar’s August 2024 fund announcement independently confirms a $235 million Magnetar AI Ventures fund, coverage of AI models, infrastructure, and applications across text, audio, and visual modalities, and a contracted CoreWeave relationship for dedicated GPU access for portfolio companies. This is a public capital-allocation and portfolio-support strategy surface. It does not establish an internal Magnetar trading model, a GenAI research lab, a dataset license, model weights, agent permissions, or AI-attributed performance. The capture note records the automatic-caption recovery and its timestamp and evidence boundaries.

Numerai / CrowdCent: title-blind agent and meta-model workflow

The title-blind queue recovered Numerai’s Out of Sample Ep 4, a Numerai Council of Elders episode whose exact upload date was not exposed by the checked metadata route. Numerai’s official company page publicly connects guest Jason Rosenfeld with CrowdCent and the Council of Elders, while CrowdCent’s About page independently identifies him as Co-Founder and CEO. In the recording, Rosenfeld describes a public 10-day/30-day crypto-prediction challenge, a CC-Points-linked meta-model, open-source model and feature tooling, an LLM-based recursive code-editing experiment, and agent-assisted dashboard/model workflows (approximately 08:12–13:27 and 19:54–25:36). The capture note records the automatic-caption provenance and identity correction.

This is public CrowdCent and Numerai-community evidence. It should not be read as evidence that the described tools, permissions, or agent workflows are Numerai’s internal production stack. A separate recovered Richard Craib interview is retained in the same note as historical corroboration; it does not change the current Numerai model, agent, or strategy record.

Two additional title-blind recoveries deepen Numerai’s historical public record. In the Weights & Biases recording, Richard Craib describes obfuscated data, temporal “era” groupings, the danger of treating rows as independent observations, feature neutralization, and a preference for residual information rather than common factor exposure (00:12–06:56; 10:03–11:29). He also discusses NLP, news-sentiment data quality, and mining company statements that are already produced at scale (17:06–18:20). The Data Driven NYC fireside, published January 31, 2022, adds a founder explanation of the tournament, Numerai Signals, stake-weighted aggregation, and the historical separation between submitted models and the meta-model (00:12–01:57; 10:59–13:53; 16:01–16:31). Both recordings are historical founder/vendor or community evidence; neither establishes Numerai’s current model inventory, agent deployment, data rights, portfolio authority, or performance. The recovery note records the local transcript hashes and boundaries.

The queue also recovered Numerai Office Hours S01E07, where a participant identified only as “ZEN” describes software-engineering experience and leading an unnamed company’s AI department (00:54–03:20). The conversation covers validation, live model changes, staking, and Meta Model Contribution (05:49–07:38; 22:27–27:28). This is anonymous community evidence; it must not be converted into a named personnel record, Numerai employment claim, or current tournament-methodology claim. See the capture note.

The recovered Epicenter 191 interview with Richard Craib, published July 11, 2017, adds historical founder evidence on obfuscated data, prediction licensing, and meta-model construction (03:29–05:00; 22:03–24:29; 24:47–30:34). It is useful for platform lineage, but it predates current AI-scientist, Predictive LLM, and MCP disclosures and does not establish current deployment, permissions, or performance. See the capture note.

The July 29, 2020 Outlier Ventures founder interview adds historical discussion of the machine-learning tournament, internal versus crowd models, model diversity, and prediction sourcing (10:03–16:33; 18:02–20:52). This is dated founder evidence, not a current model, agent, permission, or performance disclosure. See the capture note.

The distinct Epicenter Episode 348, published July 14, 2020, adds a second 2020 founder route. Craib discusses staking, model contribution, less-correlated ensembles, an obfuscated dataset described as roughly one million rows by 310 feature columns, and Numerai Signals as a way to combine external stock predictions with the platform (04:52–11:29; 19:13–21:19; 42:56–47:51). He uses social-media sentiment as an illustrative example of a weak standalone signal that might add information in combination, not as a disclosed production feature. This is historical founder media, not a current model, agent, permission, or performance disclosure. See the capture note.

The Token Summit II — Numerai recording is a distinct, short 2017 founder presentation. It covers early obfuscated-data and staking concepts plus a proposed stake-backed route to GPU/EC2 compute (00:14–06:42). The approximate dataset and staking figures are speaker-reported historical claims, not current architecture, a launched service, a benchmark, or performance evidence. See the capture note.

The separate Token Summit I panel gives a 2017 founder account of Numerai’s external data-scientist network, prediction submission, and staking incentives (05:35–07:34). Its scale and token statements are historical, speaker-reported claims; they are not evidence of current contributor counts, architecture, permissions, or performance. See the capture note.

DHF Capital / DHF Nova Fund: a claimed fully AI-managed fund and an explicit control tension

The title-blind sweep recovered Innovantage Podcast #49, published May 14, 2026, with Bas Kooijman, identified by the publisher as CEO and co-founder of DHF Capital. Its public Podbean enclosure was downloaded and processed locally with WhisperX-MLX large-v3-turbo, producing 922 automatic, non-diarized segments. The capture note records the audio and sidecar hashes while keeping the full media private.

DHF’s first-party announcement independently identifies DHF Nova Fund as a Luxembourg AI-driven multi-asset fund and describes proprietary AI systems spanning global news, market microstructure, strategy optimization, risk, and capital allocation. In the episode, Kooijman reports that the fund is “100% AI” and “zero human touch,” and describes 187 agents with roles including risk analysis, news monitoring, and reallocating from a weaker strategy to a stronger one (47:44–48:43). He also describes an optimizer reviewing a trader after every 100 trades (50:12–51:00) and a 2025 dry run using DHF’s own data (51:42–51:54).

The same account describes risk caps, automatic block levels, alerts and reporting, alongside manual oversight (49:22–50:04). That creates a useful evidence tension: “zero human touch” appears alongside controls and oversight. The source does not establish whether the phrase means no discretionary intervention, no trade approval, or marketing shorthand. Kooijman also reports a 32% efficiency improvement and a scheduled-news routine that pauses around events, checks impact, and resumes (52:00–52:53); the baseline, denominator, and independent benchmark are not provided. A separate operating-model discussion describes scheduled agent summaries, approval gates, and spending limits (40:38–42:01).

These are firm- or guest-reported claims. The public record does not disclose the 187-agent registry, model weights, training corpus, data licences, complete control implementation, customer count, audited track record, model-specific P&L, or portfolio authority. The LEI record supports the existence of a legal entity, not its technology or results. This is a dated public disclosure of a claimed AI-managed fund and an agentic operating model; it is not a ranking, validation of performance, or proof of autonomous investment authority.

Acadian: investment-AI hiring, conference agenda, and systematic-credit media

Acadian’s 2025 Client Conference and agenda PDF add a first-party event surface with an LLM session, “A Day in the Life of Acadian Data,” alpha-forecasting and portfolio-attribution demonstrations, proprietary optimization scheduling, portfolio-review tools, and an ESG analytics description involving LLMs applied to unstructured data. The agenda names Kelly Young, Brendan Bradley, Malcolm Baker, Scott Richardson, Ryan Taliaferro, Vlad Zdorovtsov, Doug Eisenstein, and Jim Soper in the associated leadership, research, investment, data, and technology roles. These are event and product-description signals; no recording, model inventory, data license, or performance attribution is public on the page.

The current VP, Investment AI Engineer role places agentic workflow work inside the Investment team. Its public description names reusable investment skills, sub-agents, human review, testing, controlled execution, data/cost/scope guardrails, collaboration with quantitative researchers and portfolio managers, and adoption/quality/time-to-value measures. It is evidence of hiring intent and desired operating boundaries, not proof of filled headcount or live deployment.

The expanded personnel and media routes include Steven Wong’s CFA Society Singapore roundtable, Andy Moniz’s CFA Society Boston conference biography, Bin Shi’s AAAIM speaker biography, and Acadian’s Devin Nial NLP post. Together they expose alternative-data/ML discussion, NLP lineage, systematic-equity ML experience, and responsible-investing text-analysis use cases. RavenPack’s Joseph Simonian video page adds a 2019 methodology route, while Evercore’s Flow of Funds catalogue lists a 2025 episode with Kelly Young and Scott Richardson covering systematic credit, data science, ML/GenAI, liquidity, and transparency. These remain dated public personnel, conference, vendor, and executive-media records—not a complete Acadian model or authority map. See the expanded Acadian source note.

The complete-feed audio reconciliation also resolves the remaining Acadian title-blind podcast routes. Episode 4 on ESG names Malcolm Baker, Matthew Picconi, and Ryan Taliaferro and describes multi-factor analysis, large security coverage, messy ESG data, and the need to fit collected information to the investment process rather than insert an off-the-shelf data set into a model (capture note). Episode 2 on the 2020 election names additional Acadian portfolio and strategy personnel but contains no verified AI, ML, agent, or model-use statement in the reviewed transcript. These recordings are retained as dated process/personnel context; neither adds a current AI deployment or performance claim.

The current Wolfe events calendar shows that this research lane continued into 2026 with the 8th AI in Finance, 3rd Canadian Quant and Macro, and 9th European Quant and Macro conferences. It also now provides the live change-detection route for Wolfe’s broader corporate-access calendar, while the supplied EventSmart event list is currently empty/disabled and the supplied event pages render as expired. Wolfe’s QES Data Feeds page adds a deeper first-party product surface: more than 3,500 stated proprietary point-in-time factors; a QES team of more than 20 quantitative researchers; and named model families including PDI/CPDI, StarPerformer, NEMO 2.0, and GINA. The QES Investable Strategies page separately reports more than 100 papers, 30+ live models, daily tracking, and monthly reporting, with model histories for MALTA, SHIELD, GEMMA, RISE, StarPerformer, LEAP, BRAIN, and GINA. These are current first-party product descriptions and reported claims; they should not be read as evidence about any named hedge fund’s internal AI stack, client adoption, model code, permissions, capacity, or independently audited post-cost results.

The otherwise title-only “From Signals to Decisions” Wolfe–Arcana page has a substantive companion in Rich Falk-Wallace’s public LinkedIn post. He describes a presentation on weekly factor moves in oil beta and hedge-fund crowding, exposure decomposition through rates expectations, software and size exposures, and FMP components, plus custom baskets and hedging solutions. This is a vendor/executive account of conference content; it does not establish client identities, live permissions, or signal performance.

XTX Markets: a named AI-lab and title-blind foundation-model podcast

The title-blind pass recovered “AI in Finance and Symbolic AI with Atlas Wang”, a December 2025 episode of The Information Bottleneck. The episode is also reachable through Riverside’s RSS feed and Spotify. This matters for discovery coverage: the show title contains neither XTX nor hedge fund, so a firm-name-only search would miss it. Local audio recovery and timestamped ASR provide a searchable internal route; automated transcription is treated as navigation evidence and not as publication-grade verbatim quotation.

In the episode, Atlas describes several distinct AI lanes in finance: workflow automation; extracting alternative signals from internet and social data; and building numerical, continuous-stream foundation models for multivariate financial time series. Around 00:49:57–00:52:39, he describes price, volume, transaction, and related market data across many instruments, with the execution strategy determining the forecast horizon, while emphasizing noise and the difference between data quantity and data quality. Around 00:55:31–00:57:22, he places XTX in the time-series-foundation-model lane. These are attributed practitioner statements, not an independent model inventory, performance result, or claim that every described method is in production.

The Stony Brook AI Innovation Institute seminar page identifies Atlas as an XTX Research Director and says he founded and leads an XTX AI Lab in New York focused on large-scale foundation models for financial time series and market data. The Princeton AI² abstract, NYU AI event page, and University of Minnesota DSI event page provide a dated public event network around foundation models, high-frequency data, scaling, causal train/test discipline, and algorithmic trading. None of these pages supplied a recording in this pass, and none establishes a complete production system.

The XTX careers page supplies a separate first-party workflow signal: quantitative research uses statistical models, machine learning, and compute for price forecasts, while exchange-trading development takes models from prototype to live trading across many instruments and includes real-time monitoring and operational controls. The public role board and AI Research Internship expose hiring intent around XTY Labs and ML infrastructure. The UT Austin profile and VITA profile provide Atlas’s public academic route, including a UIUC PhD under Thomas S. Huang and a USTC engineering degree. The OpenReview paper links an XTY AI Labs/XTX Markets affiliation to a theory paper, but does not show that its method is used in trading.

Finally, a public NeurIPS LinkedIn post supplies a dated conference and recruiting surface: Atlas describes an XTX booth, multiple accepted papers, a Generative AI for Finance workshop, and an expanding AI research team. That is personal social evidence, not a firm-wide organizational chart. The source note records the capture, timestamped ASR navigation points, unresolved recordings, and the public/private evidence boundaries. It does not include a publicly reachable document explicitly marked private and confidential.

Two further Blushing Quants episodes expand the methodology layer without creating firm-specific deployment claims. Antonio Marrazzo discusses factor validation, point-in-time data, leakage, time-aware validation, meta-labeling, and a graph-attention experiment that did not pass his tests. Nam Nguyen discusses AI-generated crisis scenarios for portfolio stress testing while retaining human judgment for unique crises. Both are useful for tracking implementation vocabulary and control boundaries; neither identifies a current fund system.

The same feed also contains a first-party AlphaBeta disclosure. CEO Oded Shimoni describes machine-learning and deep-learning models in broad systematic portfolios, dynamic factor allocation, ensemble weighting, and implementation constraints such as liquidity, borrow costs, and tradable universes. The article treats this as a firm-reported methodology statement, not as an independently audited model or performance record.

The regional title-blind pass adds Korean conference and university surfaces, a Japanese hedge-fund strategy note, a German OTHOZ interview, a French investment podcast, and European AI-for-alpha material from Etna Research. These sources widen language and geography coverage while preserving the distinction between conference metadata, educational material, vendor claims, and evidence about a named manager’s live system. See the public regional expansion note.

Singapore and Hong Kong: executive media and regional-language follow-up

The AIMA Perspectives interview with Mark Wong adds a Singapore-based, title-blind executive route for Dymon Asia. AIMA identifies Wong as Co-CEO, COO and CRO and publishes a timestamped transcript. In the 36:00–40:17 section, Wong describes generative AI as an efficiency and productivity tool with privacy and bias concerns; says a proof-of-concept began in non-investment operations; identifies research as a possible investment use case while saying more work was needed before AI-generated trade ideas or signals; and discusses exploratory demand for coding, quantitative-engineering, numerical, and data-science talent. The transcript references events around 2023, so this is historical, date-scoped evidence rather than a current 2026 deployment claim. It does not disclose a model, vendor, corpus, permissions, live authority, or performance.

The Hong Kong lane adds a new Jiemian report on Going International Asset Management / 高盈量化, with a United Daily News mirror, describing an August 2026 HKEJ / CEO AI｜EJ Tech interview with chairman Wu Chao / 吴超. The reproduced media account discusses AI-agent task separation across financial data, price/volume, news, and alternative-data research; differences in training, tuning, feature definitions, and parameter settings; hardware/algorithm/cloud coordination; and claimed university and AI-company collaboration. This is named regional-language media evidence and expands the existing Going record. It is not the underlying HKEJ recording or an independent technical audit, and it does not establish model weights, data rights, permissions, production status of every described component, or AI-attributed performance.

The associated official 高盈科技 forum page records the January 28, 2026 Tsinghua PBC School of Finance Hong Kong Forum, names Wu Chao and other academic/industry participants, and adds the company’s public framing of end-to-end AI, reinforcement learning, specialized-chip integration, and human validation. The company’s history page lists a strategic-cooperation agreement with Zhongke WenGe, while the high-frequency solution page describes deep-learning price prediction, order-book microstructure, and intraday timing. These official pages improve entity and conference resolution but remain company positioning; they do not establish a named model, partner technical scope, live fund deployment, or performance.

A separate Chinese-language AI Odyssey episode features a guest presented as a small hedge-fund founder under the alias Wang Huageng / 大厨. Its public transcript discusses price/volume regime analysis, cross-questioning GPT and Claude, and possible AI-agent interfaces to Bloomberg/Wind-style data platforms. The alias, fund, employer, model, and track record remain unresolved, so this is not assigned to Gao Ying or another named manager. See the capture note.

AIMA The Long-Short: a publisher-catalog route we had missed

The AIMA The Long-Short catalog is a useful title-blind discovery surface because the episode title often names a market, operating, or policy question rather than a firm or an AI method. The newly recovered 2026 route adds several regional and asset-management context records:

• Episode 118 with Kate Smaje, McKinsey’s Global Head of Technology and AI, is an industry-level AI strategy discussion. Local ASR now provides bounded timestamps for research as an early agentic entry point, human review near capital allocation, access guardrails/LLM gateways, and the taker–shaper distinction (15:22–17:20; 19:47–21:50; 37:38–39:23). It is not evidence of a hedge fund’s model or deployment. See the capture note.
• Episode 125 with Freddie Parker, Goldman Sachs Managing Director and Co-Head of Prime Insights and Analytics, is a sell-side view on hedge-fund performance, quant strategies, and Asia. It is not a Goldman Sachs or hedge-fund system disclosure.
• Episode 131 with Peter Kim, KB Securities’ Senior Managing Director and Head of Global & Wholesale Division, adds a South Korea market-ecosystem route.
• Episode 135 with Darren Bowdern, KPMG China’s Hong Kong asset-management tax lead, adds an operating and regulatory route for Hong Kong.
• Episode 140 with Kher Sheng Lee, Co-Head of AIMA Asia Pacific, adds a regional ecosystem, talent, and policy route.
• The Q2 recap, Episode 137, cross-links the South Korea and Hong Kong episodes and provides a compact recrawl seed. The Q1 briefing cross-links the AI episode with other current topics.

The title-blind catalogue sweep also found Goldman Sachs’ separate The Markets episode with Vincent Lin, co-head of Prime Insights and Analytics within Global Banking & Markets. Lin describes Prime Services’ hedge-fund customer base and says aggregated custody, settlement, and positioning data are used to study industry themes (00:34–01:24). He then discusses technology-sector de-risking, AI-infrastructure crowding, aggregate semiconductor exposure, factor volatility, and gross/net positioning (01:45–09:20). This is a first-party sell-side analytics account, not a hedge fund’s AI disclosure: it does not identify client accounts, disclose the aggregation methodology, or establish a named model, data-rights framework, live permission, trade, or performance result. The capture note records the recovered public MP3 and local ASR provenance.

This pass verified publisher identity, dates, guest roles, descriptions, and distribution links from the episode pages. It did not review the underlying audio or transcript for these episodes, so no additional model, data, permission, live-authority, or performance claim is made. AIMA, McKinsey, Goldman Sachs, KB Securities, KPMG, and AIMA APAC remain separate industry-context records rather than being treated as comparable hedge-fund deployment evidence.

Quiet-firm media correction: Tower and QRT

The coverage ledger had undercounted firm-controlled media for two already-researched managers. Tower’s Columbia AI-agents report, capital-markets panel report, and Tower Research Ventures AI x Wall Street report are now separately catalogued. Together they expose public language around MCP, multi-agent systems, knowledge graphs, agent harnesses, observability, zero-trust permissions, token budgets, OpenBB, MindsDB, Cohere, and research-workflow automation. They remain Tower-authored event and strategy descriptions, not model cards, transcripts, permission maps, or performance evidence.

QRT’s Technology Summit post says 130+ QRT technology colleagues attended the London event and names AWS, NVIDIA, and Snowflake as external partners. This expands the partner and firm-media route, but the post does not establish a specific vendor deployment, model, speaker roster, or investment-system connection.

The queue audit also promoted three previously captured records: Jane Street’s firm-controlled GPU/trading interview, Ernie Chan’s risk-and-portfolio-construction boundary for GenAI, and Ehsan Ehsani’s dated Crescendo Partners account of GenAI-assisted idea generation and preliminary research. These records expand the evidence map without implying that a named firm has granted an agent autonomous trading authority.

The next title-blind queue pass adds an AQR/Chicago academic lineage record, a former Balyasny PM’s detailed quantamental and domain-knowledge-graph account, and a OneChronos compute-market interview. The first two are directly relevant to research and workflow design; OneChronos is retained as adjacent market-infrastructure evidence rather than hedge-fund deployment evidence. A separate conference pass recovered the Future Alpha 2026 agenda, three dated post-event Hedge Fund Alpha coverage pages, and the upcoming CQF Institute AI/ML conference; those are tracked in the conference speaker-network draft, not treated as podcast transcripts or recordings.

Winton: machine learning, LLMs, and the research-process boundary

The title-blind pass recovered a previously absent Winton media trail. Winton’s CIO Simon Judes appears in a March 2026 Bloomberg interview, with Apple, Spotify, YouTube, Listen Notes transcript, Spreaker, and iHeart/Omny transcript metadata routes. Judes places AI alongside machine learning, big data, and cloud technology, and says recent LLM developments have been helpful; the episode does not identify a model, vendor, corpus, evaluation, or AI-attributed result.

The more specific Winton research article describes machine learning for large text-processing backtests, including a 40-year quarterly-report example involving 160,000 reports. It also distinguishes faster, data-rich research from slower, noisier settings where interpretability and simplicity matter, and explicitly discusses hypothesis-driven research and selection bias. This is firm-authored methodology evidence, not a production architecture or live-signal disclosure. The source note records the timestamps, recovered SRT route, connected 2023 Top Traders Unplugged page, and 2024 Alternative Fund Insight episode.

A full-feed refresh of Top Traders Unplugged recovered SI399 with Alan Dunne, published May 9, 2026. The publisher transcript connects AI-related capital spending and data-center concentration to possible distortions in economic and earnings data (00:01:31–00:04:05), then frames AI’s potential supply effects alongside energy and geopolitical shocks (00:16:45–00:24:00). Dunne also announces a Regime Adaptive Fund and describes a risk-budget design pairing strategic assets with trend following (00:56:38–01:05:42). This is a named systematic-investing practitioner’s macro and product-design discussion, not evidence of an AI model, generative-AI deployment, or independently verified performance. See the capture note.

The Derivative full-feed audit recovered a distinct Bernie Yu interview with Patronus Capital’s co-founder and CIO, published July 17, 2025. The publisher identifies a four-person team and Yu’s Optiver lineage, while the transcript describes S&P 500 volatility, liquidity and inventory, convexity, and a roughly two-to-four-week holding period that bridges traditional market making and investment management. This is a useful small-team market-structure and personnel record; it contains no AI or GenAI disclosure, so none is inferred. See the capture note.

The same archive recovered Oraclum Capital’s ORCA episode, published April 9, 2026. The publisher describes a survey-based system that combines crowd predictions, network analysis, and machine learning to select weekly predictors and produce directional index-options trades. Oraclum’s about page names Vuk Vukovic as CEO/CIO, Scott Alford as COO, Dejan Vinkovic as Chief Quant, and Mile Šikić as a quant-team leader responsible for AI-methodology development; its BASON page describes the Bayesian Adjusted Social Network methodology. An earlier Good Soil interview, uploaded April 27, 2022, introduces Oraclum as a predictive-analytics company operating across financial markets and elections (00:00–00:26), extending the public chronology before the fund route. These are firm and practitioner claims. The public sources do not disclose the network graph, participant dataset, model code, validation design, permissions, or independently verified performance. See the capture note.

The Derivative archive also recovers a 2024 Versor interview with DeWayne Louis and Nishant Gurnani. The chapter map explicitly covers alternative data and natural-language processing (42:58–54:26), then AI, generative models, and compliance constraints (54:27–01:01:38); the publisher describes more than 30 alpha-forecast models across multiple horizons. This is a dated firm-practitioner account and adds a distinct evidence lane to the existing Versor record. It does not disclose model names, data vendors, evaluation design, permissions, or performance attribution. See the capture note.

A separate 2025 interview with AIx2 founder Mohammad Rasouli distinguishes predictive AI for investment opportunity discovery from generative AI for due diligence, memos, LP communication, and portfolio monitoring. Rasouli’s public Stanford/McKinsey-to-AIx2 profile and AIx2’s current architecture pages add a personnel and vendor route covering sourcing, screening, outreach, due diligence, data access, and scoped agents. These are founder and vendor claims; they do not establish any named hedge fund’s adoption, model quality, customer permissions, or performance.

Current The Derivative feed: personal AI use, defense-tech data, and a negative control

The current-feed audit recovered three records that were absent from the ledger. Standpoint Asset Management founder and CIO Eric Crittenden describes using Claude, ChatGPT, and Grok for external research and data analysis and comparing outputs (46:31–48:39). This is explicit personal workflow evidence, not proof of Standpoint portfolio use, model training, or investment authority. The capture note records the transcript’s automatic-caption boundary.

Center15 Capital founder Ian Winer describes AI-driven analysis alongside space-based sensors and names Shield AI, Epirus, Shift5, and HawkEye 360 in the firm’s defense-tech investment discussion (37:32–40:56; 46:54–48:49). The route adds autonomy, satellite-RF data, and operational-technology cybersecurity to the source map. It does not disclose a Center15 financial model stack. See the capture note.

OneRiver’s Patrick Kazley interview discusses long volatility, convexity, trend following, systematic macro, and QIS integration without an AI or GenAI disclosure. It is retained as a title-blind negative control so quant or systematic vocabulary is not automatically classified as AI evidence. See the capture note.

Risk.net Quantcast: client modeling and neural calibration

The Risk.net/Quantcast feed adds a finance-native research route that is distinct from hedge-fund interviews. Pietro Rossi, an adjunct professor at the University of Bologna and senior advisor to Prometeia, describes a client-originated insurance use case for credit-rating-transition scenarios and portfolio-risk simulation (01:40–03:50). He also describes teaching a neural network to represent SPX and VIX volatility quantities and reduce calibration cost in a joint-surface workflow (15:13–20:30). The publisher links formal credit-transition and neural-calibration papers. The audio was recovered and transcribed locally; no client identity, model weights, training set, hedge-fund use, or return attribution is disclosed. See the capture note.

A second uncovered Risk.net route is the Abi-Jaber and Li episode. Li is introduced as a Morgan Stanley exotic-equity-derivatives quant strategist, but explicitly says the work was completed during his PhD before joining Morgan Stanley and is not a representation of Morgan Stanley (01:46–02:05). Abi-Jaber describes BNP Paribas sponsorship and collaboration with BNP Paribas quant Camille Illand (02:49–03:20). The episode explains a two-factor quintic Ornstein–Uhlenbeck model for joint SPX/VIX calibration and the skew-stickiness ratio, linking the measure to Greeks, P&L interpretation, and hedging (05:39–07:45; 08:29–10:46). The paired paper page supplies the formal model route. This is research-lineage evidence, not evidence of Morgan Stanley deployment, hedge-fund use, or performance. See the capture note.

The same archive adds a Barclays XVA episode with Ben Burnett and Benjamin Piau. Their discussion covers model simplification, XVA model risk, model/discounting/payout components, meta-adjustments, and calculation constraints (01:34–04:18; 19:25–23:30). Piau says the framework is used in practice at Barclays, while the final section treats AI/ML applications to XVA as an open research area rather than a disclosed deployment (25:03–25:20; 35:06–37:42). Risk.net’s publisher page provides the production context and paired technical paper. No code, model inventory, vendor, permissions, or performance attribution is disclosed. See the capture note.

The archive also exposes a Lipton–López de Prado episode that the title-filtered RSS pass missed. The guests discuss illiquidity-aware private-equity valuation, two-stage allocation, uncertainty ranges, tokenization, and a possible future quantum-optimization route (02:42–05:09; 21:45–24:28; 30:42–31:57). It contains no AI or GenAI disclosure and is retained as a title-blind quant-research negative control. The capture note records the affiliation, research, and deployment boundaries.

Risk.net’s paired Claude/agentic-tools feature and AI-and-quant-jobs feature provide an anonymous practitioner control. Risk.net says the features draw on 13 senior-quant interviews and deliberately withholds identities; the publisher index identifies Claude Code and Aristotle as tools discussed. These pages are useful for source discovery and public vocabulary, but cannot support a firm/person map or any claim about deployment, permissions, or results. See the capture note.

Man AHL: a missing CIO podcast route

Two title-blind publisher searches recovered Russell Korgaonkar’s 2023 Top Traders Unplugged episode and 2021 Trend Following Radio episode. Both identify him as Man AHL’s CIO; the older page also states overall responsibility for investment and research, Man AHL management and investment-committee membership, and an Oxford physics degree. The 2023 publisher page exposes a full timestamped transcript, and its direct MP3 was retrieved privately for provenance; the chapter map covers model aging, execution, dynamic position sizing, capacity, fees, drawdowns, and expected returns. The 2023 capture note records the transcript/audio correction. The 2021 page’s public Libsyn MP3 was also retrieved and processed with local automatic ASR for timestamp navigation; the 2021 capture note records the hash and non-diarized boundary. Its topic map covers research process, alternative data, risk limits, execution, and model iteration.

These are dated executive-media and strategy-process records. They do not disclose the current Alpha Assistant, AlphaTrend, AlphaGPT, a model inventory, training data, agent permissions, live trading authority, or AI-attributed performance. TTU123 now has a publisher transcript and privately verified audio artifact; the source notes and detailed capture note preserve the date and evidence boundaries.

Marshall Wace: title-blind quant, data, and production surfaces

The Marshall Wace teams page is a useful title-blind source because it maps Quantitative, Quantamental, Data Engineering, Platform Engineering, Production Engineering, and Software Engineering responsibilities without presenting them as an AI announcement. The Quant Dev Associate Programme connects alpha algorithms and portfolio construction to live trading, research-output verification, large-scale data processing, time-series analysis, low-latency systems, CI/CD, containerization, monitoring, and operational reliability. This supports a public description of a research-to-production chain; it does not identify a GenAI model, training corpus, agent permission, or AI-attributed result.

The public quantitative-team page names Yacine Merouchi, Hannah Moran, Bowen Tang, Mustafa Warsi, and Casper Beentjes and gives Oxford, Stanford, Imperial, Cambridge, and Oxford academic lineage routes alongside descriptions of data, forecasting, implementation, portfolio construction, live-code, and monitoring work. These are page-stated biographies and should be date-scoped; they are not a complete personnel or publication inventory.

The Capital Allocators episode with Paul Marshall is a title-blind discovery route covering firm history, quantitative and qualitative investing, portfolio construction, shorting, and “man and machines” in the publisher metadata. The linked YouTube recording now supplies an English automatic-caption track for timestamp navigation, but the captions are not manually corrected and are not used for verbatim quotation. A separate Bloomberg/ Castbox route similarly remains historical/contextual rather than evidence of a current system.

A separate London School of Economics recording provides a public, recoverable companion route from December 2020. Local timestamped ASR locates Marshall’s historical account of Alpha Capture as a sell-side virtual-portfolio and idea-database process (02:36–03:02), his discussion of processing financial statements, human intervention, and alternative data (14:03–17:23), and his description of thousands of ideas, execution-cost work, and a former Citadel algorithmic-trading hire (44:18–45:22). The recording is distinct from the premium Capital Allocators episode and from the current 2026 NBIM interview. It is useful historical founder evidence, but the local ASR contains proper-noun errors and does not establish current architecture, model weights, data permissions, agent authority, or performance. See the capture note.

One personal profile publicly connects Tudor Ciurca with a Marshall Wace risk-development role, Edinburgh AI and Bristol computer-science training, and ML/NLP and probabilistic-modelling interests. That is useful personnel-lineage evidence, but it is not firm policy or proof that the listed tools are approved or used across the investment platform. The The Org AI Engineering page is retained as an uncorroborated lead only and is not used to establish a team, lab, or named employee. The full boundaries are in the Marshall Wace source note.

Title-blind allocator and alternative-investment podcast lane

The Agentic Allocator capture note adds a previously untracked series-level route. It is not a hedge-fund-only show, but its guests expose institutional data, manager-research, governance, and AI-workflow decisions that are usually titled as allocator or investment- office conversations.

The Makena Capital Management episode provides a timestamped publisher transcript describing centralized data architecture, the transition from Excel-heavy workflows, and a staged path from structured data to AI-assisted unstructured-data access. The Aksia episode describes manager screening, co-investment underwriting, LPA benchmarking, and cross-asset intelligence around a large proprietary private-markets database. The TIFF episode describes an investment-committee knowledge layer linking historical memos to returns and exposure data. The HICO episode and SITFO episode add security, permissions, CRM integration, document workflows, and human review. The Phalippou episode adds an academic route on private-market document volume, extraction risk, and the possibility that AI-assisted manager comparison reproduces historical bias.

These are allocator, OCIO, public-fund, family-office, and academic surfaces, not proof of hedge-fund deployment. The speakers’ descriptions remain attributed, and no model inventory, permission map, independent benchmark, or investment result is inferred.

The title-blind search also recovered How I Invest E388 with Matt Wells, published June 11, 2026. The publisher biography describes Wells as a technology executive focused on AI and market-intelligence tools and identifies AlphaSense as a prior/current affiliation; that identity and employment description remain publisher-level evidence. The episode is relevant because it describes expert-call libraries, channel checks, and grounded AI as a way to connect qualitative market intelligence with quantitative investment assumptions. Wells says the workflow can aggregate expert transcripts, market research, filings, and other sources, while the conversation emphasizes citations, source verification, and analyst-built agentic workflows. The capture note preserves timestamps and boundaries. This is vendor/practitioner media evidence, not a disclosure of a named hedge fund’s model stack, customer permissions, live investment authority, or performance.

The latest Invest with AI episode, dated August 21, 2026, adds a current operating-layer account from Brett Caughran and Khe Hy. They describe a five-layer workflow vocabulary—triage, execution, decision support, rigor enforcement, and idea tracking—alongside browser automation when an investment system lacks an MCP connector, always-on local machines, adaptive research skills, and a “headless” research-management layer spanning sources such as OneNote, Slack, Bloomberg IB, Outlook, and spreadsheets. They also warn that an unconstrained 24-hour agent can produce mediocre work and large token bills, and use a roughly $20 token example against a $100,000 associate-cost comparison for extracting KPIs from confidential-information memoranda. These are practitioner examples and illustrative economics, not evidence of a named fund’s production controls or realized savings.

The feed refresh also recovered the July 17 Fable workflow episode. The speakers describe a personal eval corpus of meeting recordings, decks, email, surveys, and training material; model-generated investor skills; early experiments in codifying earnings-preview judgment; and a proposed evaluation set for MCP vendors. They also discuss making private file systems agent-legible through Markdown conversion and metadata, composing narrow subagents around complex financial models, and keeping deterministic code in the workflow where appropriate. The episode is valuable for implementation vocabulary and evaluation design, but its model-quality, cost, and financial-pattern observations are personal or conceptual accounts, not backtests or hedge-fund deployment evidence. The capture note preserves the timestamped boundaries.

The previously untracked Hudson Labs episode adds a separate finance-precision route. CEO Kris Bennatti reports an earlier test in which frontier models returned an incorrect number about 30% of the time when asked for metrics beyond four to eight reporting periods, then describes Hudson’s no-hallucination guarantee, vector retrieval and metadata, specialized guidance extraction, and a machine-learning forensic-risk score built from companies associated with fraud investigations or settlements. Hudson’s technology page independently describes multi-period retrieval, guidance identification, relevance ranking, and source-linked quotes, while its product page describes scheduled research agents and MCP access to public-market sources. The episode’s error rates, cost comparisons, backtest figures, and fraud-risk outcomes remain vendor/speaker claims: the reviewed sources do not disclose the denominators, labels, time splits, model versions, or independent audits. See the capture note.

The title-blind queue also surfaced AIAgentStore’s August 4, 2026 “Top 10 Equity Research and Financial Analysis Agents” episode. The publisher’s workflow taxonomy treats an evidence-linked research agent as more than a summarizer: it should retrieve the correct filing and period, link assertions to source evidence, update a financial model, compare results with the consensus snapshot available at the time, detect contradictions and restatements, preserve prompts/model versions/analyst approvals, and enforce material-information controls (00:00–03:36). Its proposed evaluation separates retrieval, deterministic computation, model updates, work products, auditability, and governance, with measures such as unsupported claims, reviewer correction time, consensus variance, filing-error detection, and preservation of as-filed history (20:25–27:18). This is adjacent vendor/media evidence rather than a tracked fund’s disclosure; no customer, model inventory, permission map, independent comparison, investment authority, or performance result is established. See the capture note.

A separate Zeus Capital interview with Alistair Smallwood, published May 5, 2026, adds a named practitioner route that the title-blind search did not previously contain. The publisher identifies Smallwood as Head of Applied AI at Primer and describes prior work at sell-side firms and technology-focused hedge fund Lynott Partners. In the episode he describes source/document ingestion, user-specific calculation conventions, retained company context, model updates, analyst correction loops, and a product direction that includes alternative data, monitoring, sell-side research, and expert networks (05:49–10:36; 25:32–30:45; 38:14–43:53). Primer’s February essay and March essay provide first-party author and positioning cross-checks. The record is a practitioner/product signal, not evidence that Lynott or a named customer uses Primer, and it does not establish model identity, training data, security controls, investment authority, or performance. See the capture note.

The Primer author archive also contains five distinct research-method routes: a May 15 definition of the information/tools/judgment requirements for autonomous analysis; a May 21, explicitly nonrepresentative account of more than 50 fund, asset-manager, and sell-side conversations; an April 30 proposal to record research trajectories for process replay; and July/August benchmark reports on BigFinanceBench and FrontierFinance. Together they add concrete controls for title-blind review—point-in-time source access, repeated stochastic runs, versioned answer keys, source lineage, analyst overrides, and separation of retrieval correctness from judgment. The scores, adoption observations, and benchmark-defect counts remain author/vendor claims, not independent comparisons or evidence of a named hedge fund’s deployment. The archive capture note keeps each record separate.

The same archive yielded a separate July 21, 2026 experiment on an open-ended annual-report task. Smallwood reports approximately 290 runs on a fixed Cupid plc 2011 annual report: six specialist lenses found a staff-number issue 12/12 times, while a downstream boss ignored it 12/12 times when selecting from short reports. A later pipeline developed each candidate finding before a judge compared the completed cases; the author reports that this hybrid found the staff story 12/13 times. The important signal is architectural: discovery and selection are different tasks, and majority-like aggregation can remove a rare finding before its implications are developed. These are author-reported results from a fixed harness, not a live fund deployment, independent replication, model ranking, or performance result. See the capture note.

All thirteen direct publisher enclosures have now been recovered and captured locally as timestamped JSON/TXT/SRT sidecars, covering the six initial episodes, the seven subsequent feed records, and the trailer. The local capture uses Whisper without speaker diarization, so it is a searchable verification layer rather than a quotation source. This changes the availability status of the media; it does not change the evidence boundary around deployment, permissions, model identity, or results.

The recovered audio adds several implementation details to the publisher-note layer. Aksia describes using internal transaction and document history to screen managers, compare deals by region/sector, and identify unusual LPA or PPM terms, subject to security controls. Makena describes a centralized data foundation, structured-data consolidation, and a later path for contextualized unstructured documents. TIFF describes an investment-committee workflow under development, legacy API constraints, and a continuing human role in relationship and judgment work. HICO describes discrete LPA/DDQ and meeting workflows with security, privacy, legal review, and human checks. SITFO describes manager research, document-heavy workflows, and connections to enterprise documents and email. Phalippou provides an academic governance boundary: extraction and sentiment can help process large private-market document sets, but summaries can miss footnotes or be manipulated unless the workflow preserves context. These are attributed public accounts; the capture note records the episode-level sources and limits.

The remaining seven feed records are now captured as well. Emmanuel Yimfor’s episode connects embedding-based manager matching to the risk of carrying past allocation patterns into future screening; Shaun Ng’s episode frames AI adoption around leadership, culture, governance, data complexity, and a living policy rather than a one-time technology purchase. Both are allocator/academic routes, not hedge-fund deployment records.

Paul Fleming’s episode describes an AI knowledge “brain” for retaining client, market, and firm information across silos. Alex Harstrick’s episode uses “advanced computing” as a broader operating vocabulary and discusses bespoke workflow tools, sourcing, accuracy, and cybersecurity. Al Hemmingsen’s public profile cross-links a CIO’s family-office operating perspective with public LLM skill artifacts, while his episode describes long-lived transaction/return data, browser-based coding, cloud tools, and monitoring responsibilities. These surfaces show how title-blind media can expose operating architecture without revealing a model registry.

Moonfare’s episode describes AI-assisted legal-document extraction, normalization, investment- committee memo preparation, and lead triage for a private-markets platform. Moonfare’s press archive and AI-focused strategy announcement provide first-party context, but neither the episode nor the releases disclose the internal model stack or independently measured investment impact.

The August 18 systematic-investing pass adds four source-reviewed media surfaces that broaden the comparison set without ranking firms. BNP Paribas Asset Management’s Talking Heads page features Ram Rasaratnam, CIO for Quant Equity Strategies at AXA IM Core, discussing AI and machine learning in quantitative-equity research and portfolio risk. Bloomberg Intelligence’s Inside Active page features Jeff Shen, described as co-CIO and co-head of BlackRock’s Systematic Active Equities, in a discussion of systematic active equity, alternative data, AI/ML, factor timing, and portfolio construction. Fidelity’s Alternative Angles episode with Gil Haddad has a first-party page and official transcript covering systematic equity extension, fundamental research, alternative data, and alpha capture. Thematic Investors’ episode with RV Capital and PinnBrook Capital adds a named macro-hedge-fund discussion of AI capital expenditure and labor-market risks. Local ASR recovery of the Blubrry enclosure adds a timestamped practitioner statement from RV Capital’s Ronnie Roy that the firm uses AI for ISDA-document processing, drafting, newsletters, and data tabulation, while attempting an internal valuation engine (30:19–31:07); that is operational/research-support evidence, not evidence of production trading models or performance. These sources establish public discussion surfaces and stated roles; they do not establish complete model inventories, live permissions, or returns.

A second role-title pass adds a further set of dated public surfaces. XAI Asset Management CTO Federico Fontana appears in a separate Quant Strats interview on AI’s relationship to quant research, data, reinforcement learning, and trading infrastructure. Fidelity Systematic CIO Jessica Stauth discusses earnings-call text, machine-learning tools, LLM-based sentiment extraction, regime coverage, and overfitting controls. BlackRock Systematic fixed-income portfolio manager Jeffrey Rosenberg discusses LLMs and systematic fixed income. Former WorldQuant and Third Point data leader Matt Ober appears in two title-blind interviews covering alternative-data sourcing, quant-research pipelines, data monitoring, and the transition into AI-focused venture work. A Singapore-based Evolution Exchange panel with Jiri Pik, Ernest Chan, and Jared Broad discusses bounded AI tasks, risk scenarios, strategy optimization, data quality, and explainability. These are evidence of public statements and media scope; they are not evidence of current model inventories, permissions, or performance.

The BlackRock Systematic pass now adds a second title-blind podcast route: Resonanz Spotlight’s Jeff Shen episode, dated July 2, 2025. The publisher identifies Shen as Co-CIO and Co-Head of Systematic Active Equities and describes LLMs in investment workflows plus cross-disciplinary engineering, finance, and academic teams. BlackRock’s quantitative-investing careers page separately describes 200+ Systematic professionals, academic partnerships, alternative-data inputs, and responsible AI use. These are public role, hiring, and publisher-description signals; they do not establish a complete model inventory or AI-attributed result.

The next regional sweep adds Arrowpoint founder and CEO/CIO Jonathan Xiong, whose prior Millennium Asia, GSAM, and Mellon systematic-global-macro roles make the episode useful for personnel lineage and Asia coverage. L&G’s first-party systematic-solutions page names Elena Cardella, Boyang Liu, and Raj Shah and explicitly discusses AI, index engineering, and specialist data. A Bloomberg Volatility Forum Singapore recording adds named Capula and Dymon Asia portfolio managers, while Horizon’s first-party page identifies Head of Research and Quantitative Strategies Mike Dickson in a quant-signals and AI-productivity discussion. Rayliant CIO and Research Affiliates co-founder Jason Hsu adds a China/Asia factor and machine-learning surface. The event and asset-manager records widen coverage; they do not convert regional media into firm-specific AI deployment evidence.

A full-index crawl of Flirting with Models then recovered eight additional surfaces that title-only searches had not promoted: AQR portfolio-solutions leader Peter Hecht; Macquarie QIS leader Faheem Osman; Bloomberg research-data leader Angana Jacob; Numerai founder Richard Craib; Takahē Capital’s Moritz Heiden and Moritz Seibert; Man Numeric’s Jay Rajamony; Caladan’s John Gu; and Simplify’s Roxton McNeal and Siddharth Sethi. These records add QIS, portable alpha, data-vendor, market-making, academic-lineage, and systematic-manager context. They are separately labeled because most do not disclose AI systems; the episode index demonstrates a high-recall route for finding quant personnel and workflow material that generic AI queries miss.

A separate title-blind institutional-media pass adds five more source surfaces. Campbell Managing Partner Joseph Kelly appears in a UBP-hosted discussion describing a progression from research efficiency and code completion toward agentic coding tools. AIMA’s The Long-Short identifies CFM Head of Portfolio Strategy Philip Seager in a quantitative multi-strategy episode. HFR publishes a transcript with Morgan Stanley QIS research head Stephan Kessler covering signal generation, sentiment analytics, factor characterization, and portfolio construction. Russell Investments President and CIO Kate El-Hillow is described in a manager-research episode involving neural networks, proprietary data, and simulation. Galaxy Digital research personnel Will Owens and Zack Pokorny discuss prediction markets and AI agents as liquidity or market-making infrastructure. These records broaden title-blind coverage; they do not establish firm-wide model inventories, permissions, or investment outcomes.

The August 19 title-blind pass adds three separate records. Goldman Sachs Exchanges publishes a transcript with Osman Ali, global co-head of QIS, describing language-model use, smaller-model fine-tuning for Japanese management-disclosure sentiment, and a large-scale cross-asset research workflow. The Peterman Pod links a transcript, YouTube recording, and public profiles for Nimit Sohoni, a current Cartesia AI researcher and former Citadel quant; the episode is useful for historical personnel and research-culture context, not a Citadel system claim. CFI’s FinPod publishes a transcript with Versor founder Nirav Shah describing a proprietary event database, AI/NLP use since 2018, model research, portfolio construction, risk, simulation, and AI-supported non-investment workflows. These are distinct public accounts with different source strengths; none supplies a complete model inventory, permission map, or independently attributed result.

A current Money Maze compilation, published August 20, 2026, identifies former Vantage Investment Management CEO Andrew Veglio, J.P. Morgan Asset Management Head of Investment Platform Kristian West, and Carlyle CIO and Head of Technology Transformation Lucia Soares. Recovered YouTube captions add timestamped discussion of research data foundations, approximately 7,000 daily broker reports, an agentic framework keyed to portfolio holdings, agent context and style, and enterprise AI adoption. The capture note retains the VTT and labels it automatic-caption evidence. These are named media and asset-manager/private-market control cases; they do not establish a hedge-fund deployment, model registry, permission map, or result.

The separate November 2025 Odds on Open episode identifies Versor partners Nishant Gargnani and DeWayne Louis and provides a second, non-duplicate personnel/media route. Recovered captions and the publisher description cover alternative data such as audio, footfall, and credit-card receipts; hypothesis-defined problems; supervised ML/NLP tools; model evaluation; merger-arbitrage outcomes including competing bids; and human-led research framing. Versor’s careers page adds a current 50+ staff statement and a machine-learning quantitative-research role. This is public practitioner and hiring evidence, not a model card, permission registry, or performance audit. See the capture note.

The same first-party archive also links to Versor’s April 21, 2026 announcement of an unnamed global multi-manager partnership for an AI/ML-based Event-Driven strategy. The firm describes corporate-event coverage, competing-bid exposure, dynamic exposure management, and a stated $1 billion capacity with approximately half allocated at announcement. These are firm-reported commercial statements; the release does not name the partner or disclose model inventory, mandate terms, permissions, evaluation results, or independent performance. Its AI-use disclosures nevertheless add a useful governance signal by listing data-integrity, confidentiality, copyright/trade-secret, cybersecurity, privacy, and insider-trading risks.

The archive also links to a distinct June 4, 2026 Odds on Open follow-up with DeWayne Louis. The recovered caption route describes a roughly 26-year event dataset across North America, Europe, Japan, and Australia, a constructed database joining fundamental, market, news, event, and alternative data, thousands of features, model scores for deal outcomes, and regime-sensitive exposure analysis. The timestamped account adds process detail without naming model families, vendors, feature definitions, evaluation splits, permissions, or performance results. See the capture note.

Versor’s first-party YouTube playlist, linked from its Q1 2026 LinkedIn recap, contains eight short videos across M&A, portfolio construction, alternative risk premia, outlook, and AI. The June 18, 2026 AI-use short identifies Nishant Gurnani in its description and links the full Hedge Fund Huddle episode; its caption-recovered framing describes agents as junior-researcher aids and emphasizes disciplined process integration. The short is a first-party derivative route, not an independent model or performance disclosure. See the playlist capture note.

The same playlist had seven videos marked metadata-only in the earlier pass. Public automatic English captions are now recovered for all seven. The M&A boom short connects AI activity with data-center and infrastructure investment (00:39–00:54) and frames systematic event investing as selective risk-reward analysis (01:17–01:26). The 2026 outlook opener briefly refers to AI-driven deal-risk assessment (01:03–01:18), while the Spectris/Advent/KKR and Surmodics/GTCR shorts expose structure, bidding behavior, break dynamics, regulatory scrutiny, and pre-existing holdings as the dimensions of an event-investing review (00:10–01:15 and 00:30–01:39, respectively). The portable-alpha, alternative-risk-premia, and long/short clips add portfolio, liquidity, common-exposure, stress, and index-level framing. These are first-party educational clips with automatic captions: they improve timestamp navigation and reveal how the firm packages research ideas, but they do not establish a model, dataset, deployment stage, authority boundary, or outcome. The updated capture note records the caption hashes and recovery method.

The full-feed Odd Lots audit also recovered two adjacent finance-AI episodes that title-only hedge-fund searches had not promoted. In Rob Goldstein’s BlackRock COO interview, the publisher transcript records a “first draft” workflow in which AI produces an initial document and a stated sixteen-person review checks it (09:18–11:42), an enterprise coding use case and a speaker-reported roughly 5,000-person engineering/data/modeling population (14:39–16:55), and an “open within a closed ecosystem” description in which API calls inherit user permissions across Aladdin (23:51–26:26). BlackRock is an asset manager and Aladdin provider, not a hedge-fund disclosure; the interview does not name model providers, model versions, deployment permissions, or performance. See the capture note.

The same pass recovered Noetica CEO Dan Wertman’s credit-market episode. The transcript describes semantic extraction across differently worded credit-document terms (21:01–22:23), a knowledge-graph-like database of comparable terms, and adapted open-source language models trained on proprietary data and deployed in secure single-tenant environments with additional information-extraction models (23:45–25:27). These are vendor and practitioner architecture claims. They expose a finance-specific document-intelligence pattern, not a named hedge fund’s system, investment authority, or independently validated signal. The same source note records the episode’s AI analysis of data-center financing without treating it as proof of autonomous underwriting.

A separate January 15, 2026 Odds on Open fireside conversation with Deepak Gurnani, linked from Versor’s Q1 recap, adds founder-level history: early cloud adoption during the 2013 launch period, alternative-data expansion after a 2017 conference, AI analysis of earnings-call transcripts into quantitative scores, and references to credit-card, satellite, and weather inputs. Gurnani also describes AI/ML and alternative data as part of the firm’s continuing evolution. This is a historical speaker account, not an architecture, model, permission, or performance disclosure. See the capture note.

The same personnel sweep found a Middlemark Partners webinar announcement naming Deepak Gurnani, Luke Hinshelwood, and Angelo Calvello and describing hypothesis-driven AI frameworks versus generative AI, explainability, domain knowledge, and AI as a tool versus the starting point. Versor’s Q1 recap links the route as “The Great M&A Reset & Role of AI in Equity Event Investing.” The event year is unresolved and no recording or transcript was recovered, so this remains an announcement-level discovery lead. See the recovery note.

The queue also yielded J.P. Morgan’s Market Matters episode with Pierre Chabran, recorded in April 2023. The allocator-side discussion breaks a quantitative process into data, validation, production, execution, risk, and monitoring; emphasizes point-in-time data and revision controls; and describes ML use in execution, data cleaning, and nonlinear alpha combination. It is useful methodology context, not a current disclosure from a named hedge fund. See the capture note.

The CFM follow-up also recovers a job-description and partnership lane that is more specific than the podcast summary. CFM’s public careers pages describe a newly formed AI-native quant-research team, an ML Platform team spanning data through evaluation and deployment, and an alternative-data role that names time-series ML, LLM-assisted coding, agentic AI, and prompt engineering. A historical CFM–Columbia announcement adds an academic alternative-data and economic-forecasting collaboration. These are first-party hiring and partnership signals; an open role is not proof that hiring was completed, and a historical announcement is not proof of a current production system.

The personnel/profile pass also surfaces CFM’s ML Lab. A public CFM company post names Eric Vanden-Eijnden and Anastasia Borovykh around the lab and identifies collaboration with the CFM–ENS Data Science chair led by Giulio Biroli. CFM’s own P&I interview says ML Lab members sit in research teams and bridge active academic work into the research process. A separate CFM strategy article names predictors, classification, network discovery, automated research, and extraction from text, video, and audio as use-case areas; the firm’s approach page adds ML/AI/cloud integration with 12+ petabytes of financial and alternative data, pre-deployment testing, and board override authority during extreme events. Probabl’s own financing announcement says CFM co-led its 2025 seed round. These sources make the lab, academic lineage, multimodal use-case vocabulary, and external AI-company investment observable; they do not establish the lab’s complete staffing, production models, commercial integration, or returns.

The disclosures fall into different categories. HRT and Jane Street discuss predictive-market systems, compute, and engineering constraints. Balyasny, Versor, Numerai, Point72, and Two Sigma expose research-workflow or model-development context. CFM and Voleon expose scientific and systematic-investing lineage. HFR and InfoQ are useful methodology or industry context, not hedge-fund deployment evidence.

Man Group: three recovered title-blind podcast routes

The media gap audit recovered public audio and timestamped local ASR for three Man Group routes that had previously been represented unevenly in the registry. The Tech Talks Daily interview with Gary Collier now has its Libsyn enclosure and timestamped transcript reconciled. Collier’s 2025 account covers historical ML methods, ManGPT, an open-source-first platform, Alpha Assistant, AlphaGPT, automated backtests, and human explainability. The Big View interview adds a separate Collier account of multi-model access, discretionary information extraction, document-seeded research, and data lineage. Ed Cole’s 2026 episode adds a portfolio-manager perspective on smaller or open models for discrete tasks and enterprise model-cost tradeoffs. The ASR recovery note contains the audio hashes, timestamps, and transcription limits.

These are complementary speaker accounts, not three independent audits. They do not establish a complete Man model inventory, current vendor contracts, training data, permissions, production endpoints, autonomous trading authority, or AI-attributed performance. The Ed Cole episode is a portfolio and market discussion rather than a disclosure of AlphaGPT or ManGPT ownership.

The Acadian first-party John Chisholm episode also cleared its audio gap. Its public Podbean player exposed a 48:40 MP3, and chunked local ASR recovered timestamped navigation for the historical interview. Chisholm describes early ML overfitting concerns and later incorporation of machine-learning factors as researcher expertise and tooling improved (39:16–40:17, local audio). The episode was recorded before his June 2022 retirement, so it is a dated founder account of process evolution—not evidence of Acadian’s current GenAI stack, model inventory, permissions, or returns. The recovery note records the player route, public transcript PDF, hashes, and ASR boundary.

The HRT coverage now includes the previously missed Odd Lots episode with Iain Dunning, published October 31, 2025, alongside its YouTube upload and Apple Podcasts listing. Bloomberg identifies Dunning as HRT’s Head of AI Research and describes AI for trading efficiency and short-horizon price prediction. Public transcript mirrors add attributed discussion of a move from handcrafted features toward large-scale neural models, market-event data, compute constraints, and audited risk layers between model output and orders. The transcript-derived details remain practitioner evidence; they are not a model audit. See the capture note.

No item supplies a complete model inventory, independent return attribution, or proof that an agent can place trades. The article therefore records what each source makes observable and what remains unknown.

August 29 title-blind W&B channel recovery: finance demos and research-control layers

The Weights & Biases YouTube channel was enumerated separately from podcast feeds and BrightTALK. This recovered several captioned videos whose titles do not name a hedge fund. The large-scale agentic quant-research recording shows a vendor demonstration in which a market event is routed to macro, historical-analogs, sentiment, and quant agents, followed by a synthesis agent that emits a probabilistic forecast (00:54–01:24). It exposes tool calls, token counts, cost, and latency (02:08–02:18), then demonstrates diagnosis of a missing historical-analogs result (02:18–04:26). A second workflow varies agent-weight configurations across parallel trials and uses a meta-optimizer LLM against Brier score (06:23–07:09). The capture note records the automatic-caption hash and evidence boundaries.

The same channel adds a July 2025 LG Exaone Deep market-forecasting video and a more specific January 2026 EXAONE-BI/LSEG case-study recording. The latter identifies Wonbin Ahn of LG AI Research and describes journalist, economist, analyst, and decision-maker roles that connect unstructured documents, predictions, explanations, and decisions (06:55–07:42). The publisher description names LSEG as a collaboration case study; the recording describes a forecast-score/commentary product with a four-week horizon and daily predictions (10:33–12:27). This is a speaker/publisher account and does not establish LSEG’s internal deployment, a hedge-fund customer, model weights, training data, data rights, permissions, or independently validated performance. The archive also contains a December 2025 AI-stack panel, two 2026 CoreWeave ARIA autoresearch demonstrations, a Gradient Dissent episode with Martin Shkreli covering finance software, and an NVIDIA/W&B agent-evaluation session. These materials expose vendor-described model, data, evaluation, and workflow vocabulary. They do not identify a tracked hedge-fund deployment, model weights, training corpus, data rights, production permissions, portfolio authority, or investment result. The Shkreli recording is retained as historical practitioner context, not as evidence of a current hedge-fund operation. The registry records separate the recovered caption routes from the BrightTALK pages that remain metadata-only.

The first-party W&B/CoreWeave quant-lifecycle article adds a written control-plane route. It describes research inputs such as SEC filings, earnings calls, alternative data, market feeds, internal research, and proprietary signals; dataset/code/model lineage; walk-forward and event-driven backtesting; post-trade calibration and drift monitoring; self-hosted or hybrid deployment; and LLM/RL agents operating in simulation with traceable decisions. The article explicitly frames automated trading methods as experimental and does not identify a hedge-fund customer, deployment, permission boundary, model inventory, or realized investment result.

Processed source matrix

Firm or subject	Source	Publicly observable signal	What it does not establish
HRT	Iain Dunning on AI upskilling · caption recovery note	The recovered caption layer adds timestamped navigation for Dunning’s distinction between proprietary market-prediction AI and provider-agnostic productivity assistance (03:15–03:40), plus broad role-level AI-tool adoption and compute/data-center constraints (00:31–00:45; 04:06–04:44). The productivity wording is internally inconsistent, so no numeric estimate is promoted.	First-party publisher video and automatic captions; no model inventory, data rights, permissions, current team size, or performance.
HRT	Token-spend episode	Coding assistance, experiment ideation, monitoring, agent evaluation, GPU procurement, and infrastructure constraints.	Autonomous trading or audited productivity/returns.
HRT	Odd Lots: How Hudson River Trading Actually Uses AI · YouTube · capture note	Iain Dunning describes HRT’s market-making context, large-scale neural models over market-event data, short-horizon prediction, compute/talent constraints, and risk-checked layers around model output.	Guest account and public transcript mirrors do not disclose model weights, current team size, evaluation splits, permissions, or AI-attributed results.
Man Group	Odd Lots: 86× token-spend growth · Gary Collier · Tushara Fernando · Anthropic partnership · AlphaGPT disclosure · promoted source note	The Odd Lots interview describes 86× token-consumption growth since January, use beyond technology teams, model and budget visibility, workflow ownership, reusable AI playbooks, and human/audited/risk-controlled layers around consequential actions. Man’s first-party pages identify Collier as CTO with responsibility for technology and data science across the firm, identify Fernando as Head of Data and AI responsible for implementing the generative-AI strategy, and describe Claude, Claude Skills, Claude Code, and AlphaGPT as part of the public AI strategy.	The 86× figure remains a firm-reported consumption claim with no disclosed baseline, absolute spend, user denominator, model mix, or outcome attribution. First-party strategy pages establish public remit and workflow descriptions, not model-level ownership, permissions, or return attribution.
BlackRock / Aladdin	Odd Lots with Rob Goldstein · capture note	Executive account of AI-first-draft review, coding, agent access to an enterprise platform, and permission inheritance through APIs.	Asset-manager/provider evidence; no model registry, named provider, deployment authorization, or AI-attributed result.
Noetica	Odd Lots with Dan Wertman · capture note	Vendor account of semantic credit-document extraction, comparable-term knowledge graph, adapted open models, proprietary data, and secure single-tenant deployment.	Vendor/practitioner evidence; no independent benchmark, customer identity, investment authority, or signal performance.
Systematic investing / Alan Dunne	Top Traders Unplugged SI399 · publisher transcript · capture note	AI-related macro-data distortion, data-center capital spending, regime uncertainty, and a speaker-described adaptive trend-following portfolio design.	Adjacent systematic-investing media; no AI model, model authority, live holdings, or independent performance evidence.
Patronus Capital	The Derivative with Bernie Yu · Apple · capture note	Co-founder/CIO interview covering Optiver lineage, a four-person team, S&P 500 volatility, liquidity, convexity, and a two-to-four-week market-making-style investment horizon.	No AI/GenAI disclosure, model inventory, permissions, or independently verified performance.
Oraclum Capital / ORCA	The Derivative with Vuk Vukovic and Scott Alford · Oraclum about page · BASON methodology · capture note	Public strategy and personnel record for prediction competitions, network analysis, machine learning, weekly index-options signals, and named quant/AI-methodology roles.	Firm/practitioner claims; no independent dataset, graph construction, validation split, model registry, permissions, or performance audit.
Versor Investments	The Derivative with DeWayne Louis and Nishant Gurnani · Apple chapters · capture note	Dated firm-practitioner account covering alternative data, NLP, AI/generative models, compliance constraints, global dispersion, and a publisher-reported 30-plus-model research surface.	No model registry, data rights, evaluation design, permissions, or independently verified performance.
AIx2	The Derivative with Mohammad Rasouli · AIx2 architecture · capture note	Founder/vendor route separating predictive investment AI from generative diligence and operating workflows; public architecture names sourcing, screening, outreach, diligence, and data-control layers.	Vendor/practitioner evidence; no named customer deployment, model benchmark, permissions, or independently measured outcome.
Schonfeld	Wharton FinTech interview with Tom DeBow · local timestamped capture · Miami Tech Night recap · FE AI Lab · AI technology roles	The April 2025 CTO interview describes Schonfeld GPT through Slack, email, and APIs; multiple model families; internal APIs and vector databases; coding, email, portfolio, and trade-break workflows; data curation and entitlements; an Applied Technology Team; and a firm-reported regular-use figure of roughly two-thirds of users. The separate July 2025 first-party event recap frames AI as amplifying talent and automating workflows. Later first-party FE AI Lab and hiring pages add SchonAI, pilots, evaluation, MCP, and investment-research workflow surfaces.	The interview and later pages are firm or practitioner evidence. They do not disclose a complete model inventory, training corpus, adoption denominator, evaluation results, live trading permissions, or AI-attributed returns.
Menos AI / William Wu	The Fund AI Pod EP12 · timestamped feed expansion note	The episode describes research, position-sizing, portfolio/risk, trade-reconciliation, and data-cleaning agents; a unified research hub; code-writing agents over structured quantitative data; and a product called Voice Scoring intended to examine research logic, conviction, and behavior patterns.	Vendor/practitioner claims from a platform transcript. No model card, training corpus, label construction, out-of-sample result, customer identity, or investment attribution is disclosed.
Ridgeline / Alex Benke	The Fund AI Pod EP14 · timestamped feed expansion note	Ridgeline’s Head of AI describes a system-of-record assistant, documentation-grounded retrieval, observation-loop agents, approval pauses, trade-compliance workflow support, user-like permissions, audit trails, model gateways, evaluation work, and open-model experiments.	Platform-practitioner evidence. No customer deployment list, model inventory, evaluation score, permission audit, or investment outcome is disclosed.
FactSet	The Fund AI Pod EP20 · timestamped feed expansion note	Pat Starling describes an AI Foundry spanning data, workflows, and AI research; Kate Stepp as Chief AI Officer; MCP delivery of FactSet data into Claude, ChatGPT, and Gemini; roughly 20 AI partners including Portrait Analytics and Finster AI; private LLMs; Mercury chat; model gateways; cross-model evaluation; open-model experiments; and source lineage/audit controls.	First-party executive account and episode transcript. Partner scope, data entitlements, adoption, evaluation results, and customer-specific deployment are not disclosed.
Bunch / Diana Dinis	The Fund AI Pod EP01 · timestamped feed expansion note	Bunch’s VP Product describes a three-layer sequence of clean data, governance/policies, and AI. The episode names document extraction and categorisation, human comparison against source documents, V7, standardized prompts, and LPA workflows for fees, hurdles, carry, and KPIs.	Named product-leader account from a local transcript. No independent accuracy audit, customer list, contract scope, or production-control documentation is disclosed.
Funds governance / Bernard Hanratty	The Fund AI Pod EP02 · timestamped feed expansion note	The former Citi funds executive and Irish independent director discusses board-level AI adoption, EU AI Act interpretation, and a boundary between automating workflows and replacing accountable governance roles.	Governance and industry-advocacy discussion from a local transcript; not legal advice, a fund-specific deployment disclosure, or a regulatory classification.
Hedge-fund operations / Jeb Altonaga	The Fund AI Pod EP03 · timestamped feed expansion note	The episode provides a public personnel clue connecting Jeb Altonaga’s background with Citadel and Northern Trust, then discusses RFP agents that deconstruct questions and draft responses from internal material.	Historical personnel context and practitioner/vendor workflow evidence. No current Citadel deployment, model inventory, trading authority, or investment outcome is disclosed.
BNY / Northern Trust data operations / Duncan Cooper	The Fund AI Pod EP06 · timestamped feed expansion note	The former Chief Data Officer discusses data governance, data products, and agentic workflows including knowledge-management and RFP agents.	Historical practitioner and vendor-context evidence. No named client deployment, current employer system map, or measured outcome is disclosed.
FINBOURNE / Toby Glaysher	The Fund AI Pod EP09 · timestamped feed expansion note	FINBOURNE describes a prospectus agent, MCP-routed permissioned APIs, confidence flags and human review, plus rebalancing and reconciliation agents. A customer example is described as covering about 2,500 accounts across roughly 25 custodians in under an hour with one employee.	First-party vendor account. Claimed accuracy, timing, account count, production status, and customer impact are not independently audited here; this is not evidence of hedge-fund returns.
LinqAlpha / Hojun Choi	The Fund AI Pod EP10 · timestamped feed expansion note	LinqAlpha describes a central intelligence layer for broker emails, newsletters, internal notes, research, and market data; specialist and devil’s-advocate agents; a proprietary multi-agent architecture; approximately 30 reusable agents; MCP-native integration; and coding-agent support for quantitative analytics and portfolio construction.	Vendor architecture and workflow account. No customer data permissions, model inventory, benchmark protocol, or investment result is disclosed.
LinqAlpha / Third Square Capital	LinqAlpha AI Lab · Third Square customer story · AWS architecture case study · FinAgentBench · capture note	LinqAlpha publicly describes an AI Lab organized around understanding, measuring, and building financial “Alpha Intelligence.” Its customer story describes a Devil’s Advocate Agent for source-grounded thesis challenge at Third Square. The AWS partner post names a parsing/retrieval/synthesis architecture, Bedrock Claude models, AWS storage/search components, and several LinqAlpha personnel; FinAgentBench supplies a separate finance-retrieval benchmark and author-lineage route.	First-party, customer, partner, and academic evidence are kept separate. Customer-count, speed, agent-count, and return claims remain vendor/customer claims; no independent evaluation, data-rights map, permission map, or investment attribution is disclosed.
LinqAlpha engineering and research route	DeepSeek-R1 agent repository · Hojun Choi announcement · AI Lab publication index · OpenBB integration post	The public repository documents an iterative DeepSeek-R1/Fireworks function-calling agent with structured validation, tool use, bounded iteration, and error recovery. LinqAlpha’s research index lists work on financial LLM bias, embeddings, agentic retrieval, disclosure signals, mention-market forecasting, and prediction-market risk filtering. OpenBB describes a partner API route for screening, earnings-transcript analysis, and dashboards, with an event demonstration in Singapore.	Open-source, first-party, and partner/social evidence. The repository is a demonstration, not proof of production investment use; the posts do not disclose customer permissions, model tuning, live authority, or performance attribution.
Jane Street / Arjun Guha	Jane Street Tech Talk · linked recording · capture note	A full-transcript, firm-hosted technical talk with Northeastern professor Arjun Guha on LLMs and programming languages, low-resource languages including OCaml, developer adaptation, and evaluation design. The discussion highlights the need to distinguish model failure from missing libraries, environment, or tool configuration.	External academic speaker hosted by Jane Street. It is technical and methodological media evidence, not evidence of Jane Street authorship, a live trading model, permissions, or performance.
Lumint / Alex Dunegan	The Fund AI Pod EP17 · timestamped feed expansion note	Dunegan describes Gemini for data-science prototyping, Claude Code for daily work, Cursor for model switching, GitHub integration with rigorous code review, model-generated explanations of factors, and a human approval step before sending an execution instruction.	Small currency-management firm account. It does not establish a complete production architecture, vendor contract, model evaluation, or return attribution.
Simmons & Simmons / Izzy Tennyson	The Fund AI Pod EP18 · timestamped feed expansion note	Simmons & Simmons describes its proprietary Percy AI tool, private-cloud operation, legal research and document workflows, model testing, and boundaries around human review and agentic workflows.	Professional-services workflow evidence adjacent to funds. It is not evidence of hedge-fund deployment or an investment model.
Centiva Capital / Vik Bansal	Hedge Fund Huddle — Trusted news in the age of AI · timestamped feed expansion note	Bansal describes LLM-based sentiment models, including local-language cases, while separating AI tooling from autonomous signal creation. He says a machine-generated signal would need out-of-sample checks, and describes AI-assisted code prototyping, a roughly 40% speed-up in one live-trading process after an unknown bottleneck was found, and AI-assisted log inspection during a near-outage. The panel also discusses human-led data-vendor trials, alternative data modalities, MCP/token costs, and human validation of data anomalies.	Direct practitioner and vendor-panel evidence. The episode does not disclose model names, data vendors, factor definitions, validation samples, trading permissions, or independently measured speed/return attribution.
Stoic Point / Raj Shah	Fundamental Edge — Stoic Point’s Raj Shah: AI and the Lean Hedge Fund · YouTube recording · timestamped feed expansion note	Shah describes separating deterministic Bloomberg/EQS screening from non-deterministic qualitative work in Portrait and AlphaSense; a roughly 50-screen meta-screen; MCP-connected monitoring; peer discovery; expert-network, SEC, and news monitoring; PM-process documentation; an agent-based pitch sparring step; Excel-plus-AI workflows; and a possible division between human idea generation and AI-supported risk/portfolio management.	Named hedge-fund co-founder account and public captions. No system diagram, customer/vendor contract, portfolio permission, counterfactual, or AI-attributed P&L is disclosed; captions contain automatic-caption errors.
Benzinga	The Fund AI Pod EP23 with Brad Olesen · recovered CloudFront audio · capture note · Benzinga	Olesen describes Benzinga as a financial-media and data-distribution provider, with internal corpus search and research scaffolding for reporters, customer-facing AI partnerships, a Perplexity relationship, MCP/API distribution for institutions and self-built tools, prediction-market resolution and coverage, and human review for market-sensitive content.	Named-executive interview and local automatic transcript. The episode does not disclose partner contracts, MCP schemas, model inventory, adoption telemetry, editorial error rates, or hedge-fund deployment.
Balyasny / Shu Bai	The Fund AI Pod EP21 · timestamped feed expansion note	The dated episode provides a public personnel and investment-process lineage through Barron Capital, Davidson Kempner, and Balyasny. Bai discusses underwriting AI-exposed businesses, applications, sustainable cash flow, and three-to-five-year valuation uncertainty.	Personal investment discussion, not evidence of a firm-wide Balyasny AI program, proprietary model, live trading permission, or performance result.
Balyasny / Jimmy Karalis	LSEG Hedge Fund Huddle — The search for fast and reliable data · capture note	Karalis describes Balyasny’s central data organization and catalog, taxonomy and metadata filters, code/documentation fields, trial and production-access requests, cross-strategy dissemination, data-quality arbitration, cloud-first resources, and coordination with technology, compliance, and information security around LLM tooling.	Named current data-sourcing leader and first-party transcript; no model inventory, vendor contract, permission map, data rights, deployment, or performance attribution.
HSBC Alternative Investments / Declan Sheehy	The Fund AI Pod EP22 · timestamped feed expansion note	The former HSBC Alternative Investments CTO/COO describes the importance of data pipelines, consistent data models, structured/unstructured integration, and a cross-functional intelligence layer before applying GenAI.	Historical practitioner context. The episode’s personal AUM statement and architecture views require independent verification; it does not establish HSBC’s current systems or returns.
HRT	ICML 2025: Foundation Models for Automated Trading	Official ICML abstract names Marc Khoury and describes HRT’s public framing of terabytes of market data, regime shifts, adversarial participants, latency, and foundation-model research challenges.	Official abstract only; no confirmed recording, model weights, live permissions, or performance evidence.
Jump Trading / BattleFin	LSEG Hedge Fund Huddle — Alternative data explosion · capture note	The title-blind transcript identifies Jump data-strategy leader Stewart Stimson and BattleFin CEO Tim Harrington. It covers alternative-data sourcing, mobile/IMEI, credit-card and receipt data, surveys, hiring and WARN data, weather and maritime data, dataset history, data combinations, and privacy/regulatory checks.	Named practitioner/vendor transcript; Jump is a proprietary systematic trading firm, and the episode does not establish current contracts, model inventory, permissions, or performance at a hedge fund.
Jane Street	GPUs, trading, and hiring	Compute, research-system, and technical-talent discussion from a firm-controlled interview.	Which models feed which strategies.
Jane Street / Ron Minsky and Daniel Pontecorvo	GPUs, trading, and hiring — local capture ledger · timestamped queue ledger	The firm-controlled interview describes multiple latency regimes, model placement across CPU/FPGA/GPU, specialized architectures for different data rates, fair-value prediction as one target, and researcher-throughput considerations.	A technical representative discussion, not a complete model inventory or policy. It does not identify finance-model weights, trading permissions, model performance, or autonomous order authority.
Jane Street	AI data center	Data-center and training-infrastructure vocabulary.	Model weights, training corpus, or strategy performance.
Jane Street	Signals and Threads title-blind expansion · SF Scala/Yaron Minsky capture note · Antithesis case study · BugBash agent-workflow talk	The expanded route set exposes the Hive, neural-network feature-data demand, ML-workload networking, deterministic testing, GPU-aware programming, data-center constraints, compiler/library infrastructure, and historical research/production language boundaries. Jane Street’s own Antithesis case study says the firm uses Antithesis to test the Aria distributed message bus and describes Jane Street as both a customer and investor; a 2026 Ron Minsky talk describes agents in software development alongside type systems, tests, code review, and growing interest in formal verification.	Infrastructure, validation, software-development, and historical engineering evidence; no finance-model inventory, trading authority, or returns.
Balyasny	Macro process	AI discussed within a named macro-research workflow.	Production permissions or measurable investment impact.
Balyasny	AI agents and 2026 outlook	Executive-level discussion of agents and firm operations.	Exact agent boundaries and current deployment stage.
Balyasny	Generating Alpha	Operating-model and platform context.	AI-specific implementation unless separately corroborated.
Numerai community	Modern AI research in quantitative finance · caption recovery note	The recovered community recording describes an MCP-style agent interface, fixed data and evaluation rails, a persistent computational kernel, and allowing feature engineering, model selection, and hyperparameter work inside fixed time-series and metric boundaries (02:16–06:45). It also discusses reinforcement fine-tuning of a Mistral model for code generation (04:02–04:14).	Participant/community evidence, not Numerai staff or platform-wide deployment evidence; automatic captions do not establish model ownership, permissions, capital authority, or performance.
J.P. Morgan Asset Management / Pierre Chabran	Market Matters — A Focus on Quant Strategies · capture note	The April 2023 allocator discussion decomposes quantitative research into data, idea validation, production, execution, risk, and monitoring; stresses point-in-time data and revision controls; and describes ML use in execution, data cleaning, and nonlinear alpha combination, with open-source versus bespoke-tool distinctions.	Date-scoped allocator context, not a current disclosure from a named hedge fund. It does not identify the firms behind the generalizations, model inventory, data rights, permissions, or performance.
Versor	Hedge Fund Huddle — AI at work · Odds on Open partners episode · Deepak Gurnani founder episode · DeWayne Louis follow-up · Versor Minute AI-use short · April 2026 AI Event-Driven strategy announcement · Versor first-party strategy release · Versor careers	The March 25, 2026 episode describes agents as junior-researcher analogues that can read papers, implement ideas, and run them through an internal evaluation framework while a PM or strategy lead retains the decision. The November 2025 partners episode adds hypothesis-driven alternative-data research, audio/footfall/credit-card inputs, model-based merger-arbitrage outcomes, and human problem framing. The January 2026 Deepak Gurnani episode adds founder-level history on early cloud adoption, alternative-data expansion, AI analysis of earnings-call transcripts, and AI/ML development alongside cloud infrastructure. The June 2026 follow-up adds a four-region, roughly 26-year event-dataset account, thousands of features, outcome scores, and regime-sensitive exposure analysis. The first-party June 2026 short provides a concise, firm-controlled derivative of the agent-as-junior-researcher framing and explicitly links back to the longer episode. The April 2026 release adds a firm-reported AI/ML Event-Driven strategy, an unnamed multi-manager partnership, corporate-event and competing-bid framing, dynamic exposure management, and an explicit AI-risk disclosure surface. Versor’s first-party and careers pages add ML strategy, 50+ staff, and machine-learning research hiring signals.	Interview, episode, firm-release, and hiring evidence; not a complete model inventory or independent evaluation. The sources do not establish firm-wide deployment, trade authority, data rights, partner identity, mandate terms, feature definitions, or returns.
Goldman Sachs Asset Management QIS / Osman Ali	Goldman Sachs Exchanges — Will AI Make Markets Less Efficient?	The first-party transcript describes QIS use of large and small language models, a historical progression from internally built bag-of-words sentiment models, smaller-model fine-tuning for Japanese management-disclosure sentiment, and analysis across public markets and asset classes.	Named executive account and transcript; no model names or weights, training-data rights, evaluation denominator, portfolio permissions, or AI-attributed performance.
Versor / Nirav Shah	CFI FinPod — Careers in Finance · publisher transcript · Versor research repository	Shah describes a proprietary event database, AI/NLP use since 2018, model research, portfolio construction, risk management, simulation, small initial allocations, continuous monitoring, and AI-supported code, testing, project-management, and marketing workflows.	Named founder account and first-party links; no independent model inventory, adoption measure, permission map, data-rights evidence, or return attribution. Automated transcript spelling errors are not promoted.
Citadel historical personnel / Cartesia AI / Nimit Sohoni	The Peterman Pod · YouTube · publisher transcript · LinkedIn	The title-blind episode links current Cartesia AI-research work and former Citadel quant experience, with chapters on quant/SWE collaboration, state-space models versus transformers, and AI-lab organization.	Historical personnel and public research-culture evidence; no Citadel model assignment, current Citadel affiliation, Cartesia–Citadel relationship, trading authority, or performance claim.
Hedgineer / Fundamental Edge	Driving Alpha via AI Agents in Fundamental Research	Vendor-hosted discussion describes skills for data architecture, security-master quarantine, sandboxed production clones, pull requests, observability, and DAG-style agent workflows; it also presents an anonymous $1B hedge-fund COO/CFO case involving a new data pipeline.	Vendor/customer case discussion with an anonymous client; caption transcript and publisher claims do not identify the fund, establish the stated timing independently, or prove production quality.
Hedgineer / anonymous large fund	We Got Rid of Our Forward-Deployed Engineers	Vendor-hosted discussion describes Claude Code training for an unnamed fund’s 150–200-person global engineering organization, using a portfolio-risk-engine exercise, skills, unit tests, and schema-protection hooks.	Anonymous customer example; no firm identity, attendance record, adoption measure, or production policy.
Hedgineer	Crafting an Enterprise AI Policy	The vendor describes its own AI-native operating practice: spec-driven development, agent-assisted execution and testing, AI review alongside human review, Slack-to-Linear triage, reusable skills, scheduled managed agents, and Kubernetes execution.	Vendor self-report; it does not establish a tracked fund’s internal policy, code-generation share, permissions, or control effectiveness.
Hedgineer	Gone Looping	Recovered RSS audio and timestamped local ASR describe nested loops: an inner agent-harness loop and a scheduled outer loop carrying skills and MCP servers. The worked earnings example checks covered names, pulls notes, transcripts, estimates, and consensus, produces a standard recap, updates workflow state, and separates approval-gated portfolio writes from research suggestions.	Vendor/practitioner architecture and automated transcript; no named-fund deployment, runtime audit, cost benchmark, or investment result. See the capture note.
Hedgineer	Broker Research Has an AI Problem	Recovered RSS audio and timestamped local ASR expose a permissioned skill/agent/hook/MCP repository model, the licensing and attribution problem around sell-side research, and an Aiera/AlphaSense comparison. The episode also sketches joining KPI/estimate history with broker narrative through an ontology and MCP.	Vendor commentary and automated transcript; Aiera’s first-party claims are separately sourced, and no tracked fund’s licensing, data access, live query, accuracy, or outcome is established. See the capture note.
Hedgineer	AI Orchestration: From Custom Skills to Autonomous Hedge Fund Operations	Recovered RSS audio and timestamped local ASR describe a vendor workflow from CIO shadowing and reusable skills to connectors, a consolidated data warehouse exposed through MCP, scheduled agents, usage analytics, session breadcrumbs, and an organization-wide risk-manager agent concept.	Vendor sales/process description and automated transcript; no client name, deployment contract, security review, live risk authority, or performance measure is supplied. See the capture note.
Hedgineer	Vibecoding: The Right Way	Recovered RSS audio and timestamped local ASR add practitioner detail to the publisher notes: an unnamed hedge-fund COO and people in IR, operations, and investment teams are described as building data-connected tools with Claude/Cowork and MCP; the production handoff includes dedicated repositories, deterministic data access, testing, human review, internal hosting, and permission-preserving authentication.	Vendor/practitioner account and automated transcript; the customer, code, adoption, authorization configuration, and control outcomes are not independently established. See the capture note.
Hedgineer	How Do You Hedge Against AI?	Recovered RSS audio and timestamped local ASR expose a market-risk discussion about AI as a compute-demand enabler, possible unmodeled AI exposure in beta-neutral or factor-hedged portfolios, crowding, and shared risk-model blind spots in a leveraged AI-focused-fund liquidation scenario.	Vendor commentary and automated transcript; no named-fund positions, backtest, broker data, causal attribution, or validated exposure measure is supplied. See the capture note.
Hedgineer	Kimi K3: End of the Model Moat?	Recovered RSS audio and timestamped local ASR expose a title-blind model-provider discussion about open-weight models, hosting, hardware specialization, agent harnesses, portability, and the recommendation that an asset manager own observability data, skills, memories, and structured business knowledge outside its model provider.	Vendor commentary and automated transcript; the episode-level Apple locator remains unresolved, and no model benchmark, customer implementation, or client behavior is established. See the capture note.
Hedgineer	What Does It Mean to Own Your Own Context?	The discussion treats prompts, usage traces, and workflow outputs as organizational data, using an earnings-preview example with notes, alternative data, comps, forecasts, and a PM-facing first look; it also raises retention, redaction, provider fingerprinting, and model-portability concerns.	Vendor commentary and auto-captions; no named client, provider changelog, data-rights contract, or production migration evidence.
OneChronos / market-infrastructure context	The Future of Compute Futures with Kelly Littlepage	A technical guest describes combining machine learning with classical optimization, maintaining a portfolio of algorithms, and the design problem of deliverable versus financially settled compute futures.	Market-infrastructure discussion, not a hedge-fund strategy or evidence that a compute-futures contract is live, liquid, or used by a tracked firm.
Hedgineer / Kuzu	Knowledge Graphs, Kuzu, and Building Smarter Agents	Recovered audio and timestamped ASR expose a finance-specific architecture discussion: constrained ontology extraction, front/middle/back-office lineage, SQL plus Cypher MCP servers, semantic links to gold tables and primary stores, and DSPy as a deterministic/LLM workflow bridge.	Vendor/guest architecture discussion; the audio transcript is ASR and does not establish a named fund’s implementation, accuracy, or trading authority.
Daloopa / Hedgineer	Open-Sourcing the Investor Library · Daloopa investing repository	Recovered RSS audio and timestamped local ASR describe an open skill/agent library, customer contributions, two anonymous fund implementations of a supply-chain skill, Daloopa’s Scout/Excel path, Anthropic collaboration, and press-wire parsing as a latency-sensitive data problem.	Guest/vendor account and automated transcript; the fund count, anonymous implementations, partnership scope, wire coverage, latency, data rights, and investment outcomes are not independently verified. See the capture note.
Daloopa / Hedgineer	AI in Finance with Thomas Li	Publisher show notes expose a verified-financial-data grounding thesis, an Anthropic FIS announcement, an OpenAI integration, portfolio-wide earnings analysis, and an Excel-agent use case for management meetings.	Publisher metadata and vendor claims; the accuracy denominator, partnership scope, customer identities, permissions, and investment results remain unverified.
Hedgineer	Who Owns the Last Mile? Frontier Labs Enter the Consulting Arena	Recovered RSS audio and timestamped local ASR describe the implementation layer between frontier models and domain workflows, internal AI-session cost review, cache-aware usage controls, Snowflake’s managed-agent/MCP direction, organization-wide session metadata, memory curation through MCP/CLI/skills, and a personal voice/MCP prototype.	Vendor/practitioner account and automated transcript; partnership/financing references, cost examples, customer deployments, access policies, and investment outcomes are not independently verified. See the capture note.
Basis / Hedgineer	Beyond the Chatbot with Mitchell Troyanovsky	Recovered RSS audio and timestamped local ASR add detail on deterministic business logic versus agent-facing interfaces, a “Do You Stand By This” confidence practice, scheduled mining of before/after examples to update a tone skill, internal coding-agent evaluations, and recursive context selection.	Adjacent accounting-platform discussion and vendor/practitioner self-report; no tracked-fund deployment, evaluation result, or production metric. See the capture note.
Hedgineer	The Art of Building for Agents	Recovered RSS audio and timestamped local ASR expose client rollout discussions about first-party and third-party skills, agent-oriented APIs and MCP connectors, spreadsheet/model integration, internal operations automation, custom connectors for order management and alternative data, and visible versus server-side skills for joining vendor datasets.	Vendor/practitioner account and automated transcript; no named client, skill artifact, data-license terms, access review, accuracy test, or production metric. See the capture note.
Hedgineer / Neel Somani	The Energy Behind the Intelligence · public profile · Berkeley profile · Verifiable Transformers	Recovered RSS audio and timestamped local ASR connect power-market structure, compute placement, token economics, open-model hosting, audio/video data collection, agent-assisted identity discovery, and formal methods for solver-checkable transformer properties. Public profiles cross-check Berkeley training, Citadel grid-optimization work, Eclipse, and current formal-methods/interpretability research.	Guest self-report plus public profile and paper evidence; no current fund role, proprietary dataset, production model, or investment outcome is established. See the capture note.
Hedgineer / Snowflake	AI in Finance: The Data-Centric Strategy · YouTube video and caption/audio check · Cortex AI for Financial Services · Snowflake announcement · unstructured-data strategy	The episode identifies Jonathan Regenstein as Snowflake’s financial-services AI leader and discusses data-sharing/licensing, runtime placement, evaluations, semantic layers, Text-to-SQL, and generative BI. The recovered YouTube caption/audio check adds discussion of credit-card, geolocation, and clickstream data, consumption-based access, data-quality agents, DeepEval, and semantic-layer feedback. Snowflake’s first-party materials add Cortex Agents, the Data Science Agent, Snowflake Intelligence, MCP, semantic views and knowledge extensions, and named structured and unstructured financial-data partners.	Snowflake’s product and partnership descriptions are first-party vendor evidence, and the episode is provider/practitioner evidence. Neither establishes any tracked fund’s deployment, evaluation result, data permissions, or investment outcome.
Hedgineer / Lehigh MFE	Dev Days & Lock-In Fears	Recovered RSS audio and timestamped local ASR describe an unnamed client workflow for creating first-pass fundamental research models, plus a Lehigh University Master of Financial Engineering evaluation matrix for agentic tasks that records quality, cost, and wall-clock time across harness/model configurations. The episode also describes text-based memory stores, exportability, and periodic memory condensation.	Vendor account and automated transcript; the client, evaluation matrix, task set, scores, model configurations, sample size, memory benchmark, and any fund deployment remain undisclosed. No model ranking is inferred. See the capture note.
Hedgineer	Building AI Got Cheap. Building It Well Got Expensive — S3E18 · timestamped capture note	A September 1, 2026 title-blind episode describes a vendor-internal AI spend review, session-level model/context hygiene, and a proposed mapping from inference cost to features, products, and services. It frames reusable skills, agents, MCP servers, and coordination as assets that can support future cash flows.	Hedgineer self-report backed by publisher metadata and private audio/ASR capture. No independently audited spend, valuation method, customer identity, data-rights record, production permission, investment authority, or performance result is disclosed; an ASR-only per-employee range is explicitly not promoted as a benchmark.
Point72	Matthew Granade on data science	Data-science organization and investment-research context.	Current AI organization or model inventory.
Two Sigma	Mike Schuster, The Robot Brains · caption recovery note	The recovered 2021 caption layer adds timestamped navigation for the episode introduction’s AI Core identification and a broad discussion of neural/deep-learning methods, noisy data, prediction, risk, and market-making context (00:17–00:53; 04:41–07:50).	Dated historical recording and automatic captions; no current status, model inventory, data rights, permissions, or investment result.
Two Sigma	Mike Schuster, Two Sigma first-party article · AI & the Future of Work episode 314 · YouTube mirror · caption recovery note	The recovered 2024 episode identifies Schuster as Head of AI Core, describes an episode-era team of roughly 25, and frames AI Core as a mix of engineers and modeling/research staff (03:06–03:50; 08:20–10:05). It discusses finance-versus-tech compute constraints, cloud/GPU training, TensorFlow, PyTorch, Python and Rust (10:18–13:16), and finance-specific risk limits and escalation context (15:33–17:16).	Firm article plus publisher/YouTube automatic captions; dated practitioner evidence only. No current headcount, complete model inventory, permission map, or investment result.
CFM	Jean-Philippe Bouchaud, Bloomberg	Scientific lineage and systematic-research philosophy.	A current LLM-agent deployment.
Voleon	Jon McAuliffe, Masters in Business	Co-founder/CIO describes a systematic, database- and machine-learning-based investment process; the transcript also exposes Harvard–Berkeley–D. E. Shaw–Amazon lineage. A linked YouTube automatic-caption track adds timestamp navigation around the process and execution discussion.	Current architecture, model inventory, or live performance; the caption layer is not manually verified.
Numerai	Richard Craib, Exponential View/HBR	Collective model contribution, ensembles, and the external-researcher/fund interface.	Current 7B/8B model implementation; use Numerai’s newer first-party material for that.
QIS	HFR: AI and Quant	Signal generation, sentiment analytics, and portfolio-construction vocabulary.	Hedge-fund employee evidence; the guest is from Morgan Stanley QIS.
HFT context	InfoQ: deep learning in HFT	Time-horizon choice, streaming data, overfitting, GPUs, and latency constraints.	A named fund’s production system; the speaker is a technology-company executive.
RQI Investors (Australia)	The quant edge · First Sentier Curious archive · Qingting FM Mandarin mirror · recovered Transistor player	First-party RQI and First Sentier surfaces, a newly recovered Qingting FM Mandarin mirror, publisher transcript, and local English/Mandarin ASR identify Dr Joanna Nash, Dr David Walsh, and Quin Smith. The episode describes NLP over prepared and off-the-cuff speech, corporate filings, news, and earnings calls; ML use in alpha access, portfolio construction, and risk; non-linear interpretation of analyst-forecast outliers; image/K-line pattern recognition for non-rational trading in the Mandarin route; agentic AI as a future workflow surface; and research using ML to improve existing signals and expand valuation components. The First Sentier and Qingting pages are corroborating distribution routes for the same episode, not additional episodes.	Asset-manager comparison; first-party transcript, localized summary, and automated transcripts. The public record does not disclose a model inventory, training permissions, live agent authority, or independent performance attribution. The Mandarin source supplies an additional language/platform surface but not an independent performance result. See the RQI capture note.
TD Asset Management (Canada)	TDAM Talks: Alpha Lab · recovered Simplecast RSS feed · Amazon Music mirror	TDAM identifies Julien Palardy, Philip Gendreau, and Ingrid Macintosh and describes AI/NLP over financial statements and earnings calls, millions of daily data points, faster data processing, portfolio optimization, factor design, and explicit controls against forward-looking bias in backtests. The underlying RSS enclosure was recovered and transcribed locally, adding an audio evidence layer to the published transcript.	Traditional asset-manager comparison, not hedge-fund evidence. The public record still does not disclose a model inventory, training permissions, or a named GenAI production deployment. See the TDAM capture note.
China-focused quant media	Chinatown 2.0: Robbie Yan · Buzzsprout episode · RSS · Metacast · LinkedIn lead · The Org lead	Amazon Music, Metacast, Podcast Republic, and Buzzsprout agree on the 2020 episode identity and guest description: a China-based quantitative-hedge-fund cofounder discussing trading mistakes, governance, and the Chinese quant landscape. The RSS enclosure resolves the original audio locator, but both public and signed Buzzsprout audio endpoints returned HTTP 403 during the August 28 recovery pass.	No transcript or timestamped content was recovered. LinkedIn and The Org expose personnel leads but do not safely establish that they describe the same person or a current employer; this remains a metadata lead, not current firm evidence. Retry through an authorized browser or alternate publisher mirror.
High-Flyer Quant / DeepSeek / Liang Wenfeng	36Kr Chinese interview · Lu/Caixin executive interview · English mirror · English translation and context · The Paper historical account	The 2023 Chinese interview identifies High-Flyer as an investor in a separately established DeepSeek company and says the founder viewed the AGI work as distinct from finance. The 2020 Caixin interview attributes to High-Flyer’s CEO a deep-neural-network workflow spanning data, model training, portfolio generation, and programmatic execution, with an AI Lab and internal compute investment described at that time.	Primary Chinese interviews plus secondary historical reporting. These dated sources do not disclose High-Flyer’s current financial model inventory, whether DeepSeek research is used in trading, current personnel allocation, data rights, permissions, or audited investment outcomes.
Baiont Quant / Feng Ji	Financial Times interview · Baiont first-party site · company LinkedIn	The May 2025 interview identifies Feng Ji as founder and CEO and records public claims about a roughly 30-person team, an internal foundation model, short-horizon market-data prediction, dynamic trade combination, and a stated compute-per-researcher approach. Baiont’s site corroborates its AI-driven quantitative-hedge-fund identity and computer-science/AI talent positioning.	Interview and company claims; no independent model card, training corpus, AUM audit, performance replication, data-rights record, or portfolio-permission map. The public record does not establish whether the internal foundation model is generative or whether it is used across every stage of the investment process. See the capture note.
GokuTech / 念空科技 / Wang Xiao	GokuTech about page · investment method · careers · AI-training group notice · Shanghai Securities News interview · 36Kr interview · SASR paper	The Chinese-language first-party pages identify Wang Xiao as founder partner and CIO, and expose title-blind layers for deep-learning quant research, factor-pool alpha prediction, model generalization, execution-cost monitoring, strategy deployment, and HFT systems. The May 2025 AI-training notice names three non-redacted members of the group, Jack Chen, Naruto Liu, and Erqu Qin; those names also appear in the SASR author list. The investment page describes factor generation, submodel combination, AI-model training, and backtesting. The July 2025 interview links GokuTech and a separately described AllMind research entity to LLM and multimodal-data exploration. The May 2025 36Kr interview attributes to Wang a 2017 three-person AI team, a claim that 90% of live models had shifted to neural-network/Transformer methods by 2019, and a Qwen3-based vertical model. The affiliated SASR paper lists Shanghai Goku Technologies Limited, AllMind, and Shanghai Jiao Tong University, but studies LLM post-training on reasoning tasks rather than financial prediction.	First-party and executive/media statements plus an affiliated academic paper; company infrastructure, headcount, AUM, compute, live-model share, and Qwen3 fine-tuning claims are self-reported or interview-reported. The paper does not establish a financial model, production deployment, fund permissions, or return attribution. See the capture note.
QRAFT Technologies (South Korea)	Weldon Rice — AI-Powered ETFs · Qraft/LG AI Research partnership · LQAI and EXAONE description	The dated THOR episode is backed by recovered audio and local ASR. Rice describes QRAFT’s in-house data/model workflow, machine learning and deep learning, structured-data analytical AI alongside text/sentiment-oriented generative AI, human-selected inputs with model-driven predictions, approximately intraday-to-one-month horizons, and portfolio concentration guardrails. Qraft’s first-party material resolves the ASR’s ambiguous partner reference as LG AI Research and describes EXAONE, real-time news, analyst reports, and the joint LQAI ETF.	Interviewee/company framing and automated transcript, cross-checked against Qraft’s first-party releases; no model weights, training corpus, data rights, live permissions, or independent performance attribution. See the capture note.
Timefolio Asset Management (Korea / Singapore)	J.P. Morgan — AI supply chain with Timefolio’s Singapore CIO · Brightcove player · timestamped capture note	J.P. Morgan’s first-party transcript and recovered audio identify Jae Lee as CIO/CEO of Timefolio’s Singapore business. Lee describes five Singapore multi-manager pods, a low-net fundamental Pan-Asian long-short process, supply-chain dislocation and pricing-power research, and import/export plus scraped-data checks used to shape revenue, margin, and earnings forecasts (00:01:56–00:04:18; 00:16:38–00:18:00). The host explicitly frames the episode as AI as an investment theme rather than the podcast’s usual discussion of AI in the investment process (00:00:52–00:01:46; 00:22:20–00:22:41).	First-party transcript plus recovered audio/local ASR; no generative model, AI lab, model inventory, dataset permissions, vendor stack, agent authority, autonomous trading, or performance attribution is disclosed. The data workflow is described from a discretionary/fundamental perspective.
Bridgewater AIA Labs	PAT at Interrupt 2026 · LangChain recording · alternate chapter index	A first-party case page, LangChain conference recording, and alternate chapter index identify the presenters and describe PAT as an exploratory investment-research agent using codified knowledge, LLMs, agentic workflows, entitlement-aware data access, and investor feedback.	Bridgewater-reported system and usage evidence; PAT is not presented as a trading model, and the public record does not provide audited performance or a complete model/provider inventory.
Robeco (Hong Kong surface)	Quant Street, Episode 5 · embedded Libsyn player	Robeco names Iman Honarvar as Senior Quant Researcher and Deputy Head of Next Gen Research. Recovered audio and local ASR add that Honarvar describes unstructured data such as text and social posts, NLP/ML and other statistical models, scalable cloud/data infrastructure, and a roughly decade-long foundation for collaboration across the firm.	First-party practitioner account and automated transcript; no model inventory, training permissions, production model boundary, or performance attribution. The short capture includes ASR hallucination after the substantive discussion; only the reviewed 00:00–06:50 window is promoted. See the capture note.
Robeco / Iman Honarvar	Quant Street, Episode 6 · capture note	The recovered audio and local ASR describe next-generation signals as an evolution of traditional signals using ML, NLP, and alternative data such as text, audio, social posts, and credit-card transactions; the discussion also covers earnings-call/news inputs, horizon-specific portfolios, and signal decay.	Firm-hosted practitioner interview with automatic ASR; no model identity, data license, production permission, or return attribution. Timestamped paraphrases remain bounded by the capture note.
Robeco / Jeroen Hagens	Quant Street, Episode 2 · capture note	Recovered audio and local ASR describe long- and short-horizon signals, model adaptation to changing market regimes, and incorporation of new signals over time (01:47–02:06; 02:32–02:45).	Dated firm-hosted interview with automatic ASR; no GenAI system, named model, vendor, training corpus, production permission, or return attribution.
Robeco / Matthias Hanauer	Quant Street, Episode 4 · capture note	Recovered audio and local ASR discuss quality, momentum, analyst revisions, value, multi-factor ranking, diversification, and risk- versus mispricing-based explanations (01:42–02:09; 04:30–05:21).	Dated factor-investing discussion; no ML or GenAI system is named, and no model inventory, data rights, production permission, or performance attribution is disclosed.
Robeco / Vania Sulman	Quant Street, Episode 7 · capture note	Recovered audio and local ASR describe a portfolio-construction algorithm working across a broad universe and constraints, with human exceptions for geopolitical events, controversy, corporate actions, and dynamics the model may not capture quickly (04:55–05:09; 05:20–06:48).	Dated process and model-to-human boundary; no AI model, agent, vendor, production permission, or performance attribution is disclosed.
Robeco / Vania Sulman	Quant Street, Episode 8 · capture note	Recovered audio and local ASR describe factor ranking, an in-house portfolio-construction algorithm, risk/cost/liquidity/turnover constraints, client sustainability constraints, and an in-house risk model separating static from dynamic exposures (02:19–03:06; 03:32–05:20).	Dated firm-hosted process disclosure; no named model family, vendor, training corpus, agent, production permission, or AI-attributed performance is disclosed.
Robeco / Jeroen Hagens	Quant Street, Episode 9 · capture note	Recovered audio and local ASR describe risk constraints, multiple signals, benchmark awareness, and customization by risk-return profile, universe, and sustainability preferences (02:18–03:17; 02:52–03:39).	Dated firm-hosted interview with automatic ASR; no named model, vendor, training corpus, agent permission, or return attribution.
Robeco / Harald Lohre	Quant Street, Episode 1 · Libsyn player · capture note	Public audio recovered and locally transcribed. Lohre describes a dated roughly 20-person quant-research team, documented and transparent research steps, and ML inspection through nonlinearities, asymmetries, feature importance, and interactions (02:08–02:45; 03:20–04:08).	Firm-hosted interview with Dutch/English automatic ASR; no named model, vendor, training corpus, permission, current headcount audit, agent, or performance attribution.
Robeco / Harald Lohre	Quant Street, Episode 11 · capture note	The recovered audio and local ASR place human expertise around investment-thesis design, model oversight, performance and trade monitoring, explanation, and checking model relevance; Lohre also discusses confirmation bias and prediction/outcome feedback.	Firm-hosted practitioner interview with automatic ASR; no current GenAI system, agent permission, or investment-performance disclosure. Promotional or comparative wording is excluded from synthesis.
Robeco / Jan Sytze Mosselaar and Dijana Kostic	Emerging markets, concentration risk and the evolution of quantitative investing · capture note	First-party transcript from a title-blind route covering emerging-market quant research, proprietary signals, model evolution, NLP/ML and alternative-data language, a research-and-approval funnel, and human handling of data and implementation constraints.	Dated practitioner statements; no model inventory, training corpus, feature definitions, live permissions, production endpoint, or independently audited performance.
Robeco (Hong Kong surface)	PodcastXL: The pursuit of alternative alpha · recovered Libsyn player	Published transcript, recovered audio, and local ASR name Weili Zhou, Daniel Ernst, and Mike Chen and discuss AI/ML definitions; geolocation, satellite, credit-card, text, and audio data; economic and fundamental checks before adopting alternative datasets; data-provider persistence and research infrastructure; corporate culture and stakeholder sentiment; human oversight; and a human-versus-machine report-rating experiment.	First-party transcript plus automated audio layer; the publisher warns transcript accuracy is not guaranteed. No model inventory, data rights, permission map, live authority, or performance attribution is disclosed. The repeated end-of-file ASR tail is excluded. See the PodcastXL capture note.
Pictet Japan	Japanese quant-AI article · original Japanese report	Pictet identifies David Wright as Head of Quantitative Investments and publishes a June 2026 report. Its public research describes non-linear ML models, a news-based forecast-revision-momentum test against IBES, an almost-zero incremental result after earnings-related news was removed, rising model complexity, and gradient-boosting analysis finding that 60–70% of recommendations were attributable to monitored linear factors.	First-party Japanese research artifact with an attributed translation note; the findings are not independently reproduced here and do not disclose a full model inventory, permissions, or live trading authority. See the Pictet capture note.
Robeco China	Quant Street Episode 5 — Simplified Chinese	The Simplified Chinese localized page names Iman Honarvar as Deputy Head of Next Gen Research. The shared embedded audio is now recovered through the Hong Kong player and provides an English timestamped ASR layer for comparison.	Localized publisher summary, not a translated transcript; compare the page variants for translation drift before promoting localized wording.
Robeco Japan	Quantitative investing capability page	Japanese firm page exposes AI/alternative-data framing and named quant personnel across research and portfolio roles.	Capability and personnel evidence; titles do not prove model ownership, production permissions, or performance.
Evolution Exchange Singapore	AI in Hedge Funds: Practical Integration for Portfolio Managers · Apple Podcasts	Recovered Acast audio and timestamped local ASR add practitioner detail to the publisher summary. The speakers describe task decomposition across data processing, failing-idea identification, scenario analysis, risk, coding, portfolio monitoring, and strategy testing; they mention MCP-connected calculation tools, a portfolio-correlation agent, a live-monitoring agent, “corrective AI” that flags possible trade errors rather than making the trade, cost/data-quality/reliability constraints, and a deliberately non-visible AI layer inside normal workflows.	Guest/vendor discussion, not evidence of a specific fund’s production deployment or results. The transcript is automated and the examples are not independently audited; no customer identity, model inventory, data rights, permission map, or performance attribution is disclosed. See the capture note.
Infinity Global Asset Management / Jongwon Roh	Korea Investment Week 2026 program · replay/material locator	The Korean/English conference program lists Roh as CIO and names a session on the next generation of Korean hedge-fund management through AI and quantitative strategies. The same program exposes Robeco’s APAC quant-client-portfolio-manager session on AI and uncorrelated alpha.	Official program and session metadata; replay/material access is gated. No transcript, slides, model inventory, deployment evidence, or performance attribution was publicly accessible in this pass.
Asset Management One / AMOAI	Japanese Hedge Fund Eye Vol.45	The May 2026 note by Kohei Hayashi explains statistical-arbitrage process design and publicly maps NLP to text features, ML to signal creation/combination, reinforcement learning to execution and portfolio construction, and deep learning to high-dimensional inputs such as satellite imagery.	First-party Japanese strategy note; it does not disclose a complete model inventory, model weights, training data, permissions, or independently attributed returns.
OTHOZ Capital / Julien Florian Jensen	Hedgework Talk · Podigee episode · German ASR recovery note	The German publisher identifies Jensen as Executive Director. In the recovered 2023 audio, he describes OTHOZ as a Berlin-founded technology group with investable AI products since 2018; three activity areas spanning AI-assisted equity research, index enhancement, and quantitative AI-driven asset management; strategy-specific proprietary ML models; point-in-time data storage and substantial data cleaning; ensemble/metamodel construction; and weekly target-portfolio generation followed by portfolio-management checks (00:00–17:00).	Dated guest account translated from automatic German ASR; no model names, weights, training corpus, vendor contracts, permissions, current continuity, or independently verified product results are disclosed.
Etna Research / Marco Jean Aboav	Investology episode · Etna first-party site · Aboav’s public post	The episode identifies Aboav as an ex-hedge-fund manager and Etna CEO and covers data scouting, data engineering, quant pods, alpha discovery, and adaptive strategies. Etna’s site names Aboav, David Ko, and Davide Panelli and describes frontier AI for institutional clients and asset owners; Aboav’s public post emphasizes data quality before AI.	Public interview, company page, and social-post evidence. Product, speed, and cost claims remain company claims; no client identity, model inventory, benchmark, data right, or return attribution is disclosed.
France / Ploovers / Franck Béon	Les Investisseurs 4.0 · French ASR recovery note · Ploovers team	Recovered French audio names Millennium Capital Partners as a prior quantitative-hedge-fund employer (07:15–07:39) and describes algorithmic/arbitrage work, market-data and reference-database construction, and ETF-relative alpha analysis using factor models, Fama–French-style models, and Random Forest (05:08–07:39; 17:32–19:02). Ploovers’ first-party team page identifies Béon and Marc Rousseau as co-founders and describes their institutional backgrounds.	Timestamped guest account plus first-party/secondary biography corroboration; no current hedge-fund operation, model architecture, training data, live authority, permissions, or independently measured result is disclosed.
Korea University Financial Technology Lab	Official lecture index · LLM investment analysis · AI versus financial analysts · ML in asset management · caption recovery note	The lab’s public Korean/English index exposes title-blind academic media on LLM investment analysis, AI versus analysts, machine learning in asset management, portfolio optimization, scenario analysis, and backtesting. Korean automatic-caption tracks were recovered for all three videos, with timestamp coverage recorded in the capture note; no English translation was exposed.	Academic teaching/research media, not evidence of a named fund’s deployment, client data, or investment result. Platform-generated captions still require translation and speaker/content verification before substantive synthesis.

| Kinea Investimentos / Kinea Insights | Kinea first-party agent essay · Kinea Insights episode · Kinea team page | Kinea’s Portuguese first-party media discusses autonomous agents, tool use, productivity, business-model exposure, and how those developments inform the positioning of its multimarket funds. The public team page names Mariana Marques Smidt as Head of Data Science, Rodrigo Zobaran as Head of Quantitative Research who coordinates AI-infrastructure initiatives, and Ruy Alves as a multimarket co-manager. | First-party investment commentary and personnel evidence. It does not disclose a complete internal agent inventory, model permissions, training data, production authority, or independently attributed investment outcome. | | Kinea / Ruy Alves | Os Economistas episode · Market Makers episode on DeepSeek · XP Expert Talks write-up and video | Portuguese-language practitioner and allocator media identifies Alves as a Kinea multimarket manager and discusses AI-market structure, DeepSeek, automation, valuation, and the implications of AI for investment positioning. | Public interview and publisher summaries; they do not establish Kinea’s internal model inventory, data rights, live decision authority, or performance attribution. | | Itaú Asset Quantamental | Itaú Views episode · Itaú quantitative-competition release | The Portuguese episode names Victor Dweck, manager of Itaú Asset’s Quantamental fund family, and Pedro Barbosa, Fund of Funds director, in a discussion of quantitative funds, investment robots, AI, and human capital. Itaú’s first-party release documents a student quantitative-robot competition and describes a finalist project using machine learning. | Asset-manager and talent-pipeline evidence, not proof of a specific live model or GenAI deployment. The release does not disclose model governance, data permissions, production ownership, or independently measured results. | | Noax Global / Ivan Blanco | Zona Quant episode · recovered Ivoox enclosure · capture ledger · Spanish fund article | Recovered audio and local Spanish ASR identify Blanco’s quantitative background, factor-investing framework, Noax’s systematic equity/volatility/fixed-income components, sector diversification, and a human fundamental-review step after systematic selection. The separate Spanish article describes Noax as combining factor investing with machine learning and AI. | Dated practitioner account plus secondary publisher reporting. The sources do not establish current Noax status, model architecture, training data, permissions, or current performance; historical return language is not promoted as validated evidence. | | Neo Ivy Capital / Renee Yao | Odds on Open episode · SEC adviser record · Neo Ivy public profile · public AI article | The episode identifies Renee Yao’s former Citadel and Millennium roles and presents Neo Ivy’s self-described AI/statistical-arbitrage process. SEC and LinkedIn surfaces independently resolve the adviser/entity identity; LinkedIn describes statistical arbitrage, deep learning, and large-scale parallel computing, while the public article is authored by Zhilei Xu, Yangfan Li, and Renee Yao. | Podcast, regulatory, company-profile, and company-hosted research evidence. Promotional AUM, “self-learning” descriptions, model details, and performance implications remain unverified; no current production architecture or permission map is disclosed. | | Equity Data Science / Sandeep Varma and Benjamin Lieblich | Momentum episode · EDS leadership · Sandeep Varma profile · EDS event archive | The February 2026 episode and EDS pages connect Varma’s Bridger/Herring Creek quantitative and fundamental background with a platform that joins research, portfolio construction, risk, attribution, and AI. EDS’s event archive adds public discussions of AI-driven risk management, research-management systems, and quantitative workflows. | Vendor, founder, and event evidence. EDS does not identify customer-specific implementations in the reviewed sources, and the material does not establish a tracked fund’s deployment, data rights, model evaluation, or investment outcome. | | Wright Research / Sonam Srivastava | The India Opportunity episode · Wright first-party site · Wright portfolio overview · Wright AI/ML fund material | Indian quant/PMS media and first-party materials identify Srivastava as Wright’s founder and portfolio manager, with prior HSBC, Edelweiss, and Qplum experience. The public materials describe factor and regime modeling, AI/ML forecasting and allocation, and a workflow combining automation with human oversight; the team surface also names CTO Vinod Reddy Kotha and machine-learning adviser Dr Miquel Alonso. | First-party product and biography claims plus podcast metadata; the reviewed record does not establish model architecture, training data, permissions, live decision authority, or independently audited performance. Product marketing figures are not promoted as verified results. | | JioBlackRock / Rishi Kohli | The Brand Called You episode · JioBlackRock SAE page · SEBI personnel addendum | Transcript recovery identifies Kohli as CIO and records his description of BlackRock data-driven technology plus Jio distribution (08:06–09:28), Aladdin’s integrated data/model/trading-control workflow (10:50–12:44), a local/global team mix (13:10–13:57), and a fully systematic/quantitative objective (15:35–16:56). JioBlackRock first-party pages separately describe SAE, machine learning, alternative data, Aladdin, and human oversight; SEBI records Kohli’s CIO appointment effective August 4, 2025. | Speaker and first-party product evidence; it does not establish a complete model inventory, training/data-rights map, production permissions, or independently audited investment outcome. | | FINANZ26 / Pictet, BWM Value Investing, and WealthArc | FINANZ publisher episode · FINANZ26 event metadata · capture note | A title-blind Swiss panel names Reda Jürg Messikh (Pictet Quantitative Equities and Solutions), Georg von Wyss (BWM Portfolio Manager), and Dennis Hagander (WealthArc). The recording covers LLM-assisted research/summaries/macros with human verification (11:30–13:38), ML signal construction and a speaker-reported internal experiment on cross-data interactions (09:39–10:35; 14:02–14:55), specialist-integrated raw-data/compute workflows and a corporate LLM/agentic-AI body (19:47–21:50), four-eyes review/no automatic trade handoff in Messikh’s account (33:34–34:55), and WealthArc’s lineage/governance/data-quality framing (15:15–17:04). | Adjacent buy-side and wealth-technology discussion; it does not establish any firm-wide model inventory, training corpus, permissions, production deployment, or independently audited return contribution. | | Applied AI / Stefan Jansen | Macro Hive episode 365 · Stony Brook seminar abstract · capture note | A July 3, 2026 transcript-bearing interview with Applied AI’s founder and CEO describes LLMs for hypothesis/feature variation, constrained agents that call deterministic tools and emit standardized research artifacts, RAG/provenance/knowledge-graph patterns, execution-focused reinforcement learning, and human goal-setting. The public book/seminar material adds six supporting libraries and a workflow spanning data, diagnostics, models, backtesting, and live operations. | Methods and implementation-vocabulary comparator. The speaker’s former investment employer is unnamed; the sources do not establish any client deployment, model inventory, permissions, live trading authority, or independently audited result. | | Qode Advisors / Rishabh Nahar | PMS Bazaar transcript · The India Opportunity Show · Money TV MPU playlist · Qode first-party site | A dated India-facing manager interview identifies Nahar as Qode’s Partner and Fund Manager and describes a self-taught coding path, evolving data-driven models, macro and stock-level research, and a stated preference for alternative data when asked about emerging quant-investing trends. The Money TV playlist now resolves three relevant Hindi video/caption routes: MavenArk’s AI-driven wealth-technology presentation, Savart’s AI-supported investment-advisory presentation, and Wright Research’s quant/robo-adviser presentation. The Qode video remains a separate episode-level identity route and is not substituted for the PMS Bazaar transcript. | Named-manager testimony, publisher transcript, playlist metadata, and automatic Hindi captions; no model code, training corpus, data rights, evaluation design, production permissions, or independently verified performance. Self-reported AUM and performance statements remain source-local claims. | | SpringPad / Pratik Chakraborty and Rahul Chandra | Hindi-facing AI workshop page · alternate event page · first-party team page · capture note | A regional search recovered a Hindi-facing “Stock Market Using AI” workshop and first-party instructor pages. The public material names AI-assisted market research, risk evaluation, backtesting, and automated workflows, while a public attendee post names tools discussed in the workshop. This expands India-language and role-title coverage even though the event is not evidence of a hedge fund’s internal deployment. | First-party marketing pages and attendee testimony; the recording/Hindi transcript remains unresolved. No claim is made about tool quality, live trading authority, or performance. | | Saudi Exchange Algorithm-Enhanced Trading Fund / Saudi Fransi Capital and Winton | Arabic fund prospectus · capture note | A Saudi Exchange prospectus publicly describes a systematic Saudi-equity fund whose algorithm program covers return prediction, risk prediction, portfolio construction, and cost control. It assigns Saudi Fransi Capital the fund-manager role and names Winton Capital Management Limited as investment adviser; Winton’s stated responsibilities include developing and operating the algorithm program, processing data, assigning target portfolio weights, and updating the program. | Regulatory prospectus evidence of a documented operating design, not proof of live scale, model performance, training corpus, or current implementation beyond the prospectus terms. The document does not disclose model weights, specific vendors, or the underlying data sources. | | Pictet Asset Management / David Wright | Monetary Matters episode · Pictet Japanese research note | A second publisher surface identifies Wright as co-head of Pictet’s quantitative investments and describes a process using many decision trees/gradient boosting, more than 400 features, and a roughly 20-day relative-performance horizon. The episode frames generative AI as a research/tooling question distinct from the forecast models described; the Pictet note supplies the first-party research context. | Publisher and first-party strategy discussion; the episode’s product and performance language is not independently audited here. No complete model inventory, training permissions, or live authority map is disclosed. | | PGIM Quant Solutions / George Patterson | Top Traders Unplugged episode · canonical recording · capture note | The June 2026 episode identifies Patterson as CIO of PGIM Quant Solutions. The recovered recording adds a title-blind discussion of regime detection, portfolio construction, machine learning, and language models as a source of market data and research signal for quantitative equity and multi-asset work; it also describes tracing positioning back to raw data (00:32:29–00:34:35; 00:33:13–00:33:32). | Dated practitioner account with automatic captions; it does not establish PGIM’s current model inventory, data rights, production permissions, autonomous trading authority, or investment outcome. | | Macquarie Asset Management / Systematic Investments | Macquarie systematic-investing article · SB Talks episode | Macquarie’s March 2025 first-party article describes its global Systematic Investments team, active quantitative equity strategies, a library of more than 1,000 proprietary signals, cloud and Python research-infrastructure changes, factor-calibration models, and AI/ML use in strategies and risk management. The February 2025 episode identifies Ben Leung as Head of Systematic Investing and adds a publisher personnel route for structured signals, investor-behavior inputs, AI/ML, and risk management. | First-party strategy page plus asset-manager interview; no model architecture, vendor, training data, permission map, production endpoint, or independently attributed result is disclosed. Macquarie Asset Management, Macquarie Bank/QIS, and regional legal/distribution entities are kept separate. | | RQI Investors / First Sentier Investors (Hong Kong vehicle) | Hong Kong launch release | The September 2025 first-party Hong Kong release describes the RQI Global Value Fund launch for Hong Kong retail investors, calls it the latest vehicle for RQI’s Global Value Strategy, names an AI-enabled Alpha Signal Overlay, and identifies Dr Joanna Nash, Dr Ron Guido, Dr Wang Chun Wei, and Dr David Walsh. The release says the Hong Kong material is issued by First Sentier Investors (Hong Kong) Limited and has not been reviewed by the SFC. | First-party product, personnel, and distribution evidence; performance language remains source-local. It does not disclose model inventory, data rights, live authority, independent performance attribution, or whether the overlay is used identically across other RQI or First Sentier vehicles. | | BlackRock Systematic / Jeffrey Rosenberg | FICC Focus — Credit Crunch episode | The June 2026 episode identifies Rosenberg as Managing Director and Senior Fixed Income Portfolio Manager for BlackRock Systematic and discusses LLMs, systematic fixed income, rates decorrelation, and a systematic ETF. The public discussion is relevant to systematic-investing workflow and product surfaces even though the title does not use “hedge fund.” | Podcast metadata and guest commentary; no internal model inventory, data permissions, deployment authority, or independently measured AI contribution is disclosed. | | JCube Capital Partners | JCube first-party firm page | The Singapore firm describes itself as a MAS-licensed systematic fund manager, with statistical and empirical models, machine-learning signals, alternative data, and large-language-model outputs included in its public research framing. It also describes work with accredited and institutional investors and VCC structures. | First-party positioning and licensing statement; the reviewed page does not disclose fund names, model architecture, training data, data rights, or independently measured outcomes. Verify licensing through the MAS directory before treating the statement as a regulator-confirmed fact. | | Aggregate Asset Management / David Toh and Eric Kong | Machine learning and the art of investing · 2026 AI-investing announcement | Aggregate’s first-party material describes a machine-learning process for stock selection and names the people involved in developing the fund process. The later announcement says the firm is expanding its proprietary “Deep Deep” model in its flagship fund. | Firm-reported strategy and personnel evidence; no model weights, evaluation protocol, training corpus, production permissions, or independent performance attribution is disclosed. | | Avangard Investments / A.L.F.R.E.D. | Avangard strategy page | The Australian manager describes A.L.F.R.E.D. as an adaptive learning system for ranking equities and derivatives. The page says outputs are reviewed by the investment team, positions are traceable to data and rules, and the Avangard Systematic Australian Equity Fund began 1 July 2026. | First-party system and fund description; no model implementation, data rights, independent performance record, or GenAI-specific component is disclosed. The stated human review boundary is retained. | | STANLIB Systematic Solutions / Chetan Ramlall | Moneyweb interview · Citywire podcast mirror | Ramlall is identified as Head of Quantitative Research. The transcript discusses machine learning, alternative data including satellite/drone imagery, repetitive-work automation, risk quantification, and the changing skills required in fund management. | Publisher interview and guest account; no STANLIB model inventory, production authority, training data, or independently measured investment result is disclosed. | | NMRQL Research / Stuart Reid | TWIML episode · recovered audio | The recovered 2018 audio identifies Reid as Chief Scientist at NMRQL and gives a dated practitioner account of deep-learning market prediction, regime shifts, online/incremental learning, interpretability, and performance monitoring. Reid also reports approximately 2,000 neural networks in production in NMRQL’s funds at that time (11:16–11:52). | Historical speaker account and recovered ASR; the figure is not independently audited. The source does not establish current NMRQL status, current personnel, model versions, permissions, live results, or a current GenAI/LLM program. |

Follow-up pass — August 17, 2026

Subject	Source	Publicly observable signal	What it does not establish
Sparkline / Kai Wu	AI + NLP and intangible value	GMO background, a GMO-linked quant-fund spin-out, and use of ML/NLP and alternative data to quantify intangible-asset pillars.	Model architecture, permissions, or performance.
Stoic Point / Raj Shah	AI and the lean hedge fund	Screening, research, monitoring, deterministic versus non-deterministic screens, computer use, monitoring, and pitch review are named as workflow components.	Independent verification of the workflow or its investment effect.
Epoch / Garret Brennan	Deterministic AI for institutional quant workflows · Epoch launch post · timestamped capture note	Brennan describes an infrastructure-first platform in which an LLM parses research requests while a traditional C++ backtesting/runtime layer performs computation; specialized agents communicate through an internal language and funnel into a broader interface. The launch post adds global and alternative-data coverage, deterministic backtesting, named data-provider relationships, and a company claim that a small group of firms stress-tested real research workflows.	Vendor and speaker claims. No customer identities, contract scope, model inventory, training corpus, data rights, permission map, autonomous trading authority, or independently measured financial outcome is disclosed.
OneEye Capital / Eren Biri	AI-native hedge-fund discussion	In-house options data and AI/ML vocabulary around calibration, regimes, optimization, and empirical pricing.	Independently validated model inventory or returns.
OneEye Capital / Eren Biri	Blushing Quants episode 28 — audio-backed capture · capture ledger	The publisher identifies Biri as OneEye’s founder. The publisher MP3 was recovered and locally transcribed; timestamped windows cover a tech-focused volatility/options process, discretionary overlays, scenario-based risk framing, ensembles of signals, and a distinction between LLM-assisted research and latency-sensitive execution.	Automatic ASR is not diarized or word-for-word audio verified. No model inventory, data rights, permission map, or performance attribution is disclosed.
Quant research / Denis Lukyanov	Blushing Quants episode 17 — GenAI agents and trading systems · capture ledger	Lukyanov separates programmed quantitative logic from GenAI orchestration. The recovered publisher audio and local ASR cover bounded data collection, preprocessing, synthesis, critic agents, multi-model checking, graph exploration, observability, forward testing, and retaining trading authority in explicit logic rather than an LLM.	Automatic ASR is not diarized or word-for-word audio verified. The episode does not establish a current employer, fund deployment, model configuration, or independently measured result.
Historical Citadel lineage / Jerome Busca	Blushing Quants episode 31 — Citadel, alpha decay, and quant infrastructure · capture ledger	The publisher and recovered audio identify historical Citadel hedge-fund experience and discuss data infrastructure, alternative-data exploration, backtest overfitting, and AI-assisted research speed.	Historical personal account only; the local ASR is automatic and not word-for-word audio verified. It does not establish Citadel’s current systems, personnel, permissions, or investment results.
Chicago Global / Ben Charoenwong	Blushing Quants episode 24 — academia, hedge funds, AI, and applied finance · capture ledger	Charoenwong’s public account describes mid-frequency and ex-US work where alternative data may help, monitoring of foundational tabular models, LLM-assisted structuring of unstructured data, simple/tree-based models, feature engineering, and a preference for deterministic signal inputs to preserve reproducibility. The publisher MP3 is now locally transcribed with timestamps.	Personal practitioner account; automatic ASR is not diarized or word-for-word audio verified. The episode does not identify model versions, data vendors, training corpora, production permissions, independent results, or a complete Chicago Global system map.
Quant research / Antonio Marrazzo	Blushing Quants episode 33 — factor research with data and ML · capture ledger	Marrazzo discusses point-in-time data, realistic trading frictions, survivorship and look-ahead controls, graph-attention experimentation, feature importance, classification targets, recursive windows, barrier labels, meta-labeling, overlapping-sample handling, and time-aware validation with embargo. The publisher MP3 is now locally transcribed with timestamps.	Named practitioner methodology, not current-firm evidence; automatic ASR is not diarized or word-for-word audio verified. The referenced paper and all performance claims remain unverified; no employer, model inventory, data entitlement, or live permission is established.
Quant research / Nam Nguyen	Blushing Quants episode 34 — sell-side/buy-side quants, Monte Carlo and AI · capture ledger	Nguyen provides sell-side derivatives/model-validation and buy-side research context, then discusses AI-generated crisis scenarios for portfolio stress testing, the uniqueness of crises, human intervention, and possible market-structure effects from AI-assisted code and strategy production. The publisher MP3 now has a local timestamped ASR pass.	Industry-methodology discussion with no current employer or named fund system. Automatic ASR is not diarized or word-for-word audio verified. It does not establish an implemented scenario generator, model inventory, permissions, or investment result.
AlphaBeta / Oded Shimoni	Blushing Quants episode 25 — low-correlation strategies, research, and ETF innovation · capture ledger	Shimoni publicly describes AlphaBeta’s use of statistical, machine-learning, and deep-learning models in broad systematic portfolios, dynamic factor allocation, ensembles, point-in-time data checks, and implementation constraints including liquidity, borrow fees, locates, and tradable universes. The publisher MP3 now has a local timestamped ASR pass.	First-party executive methodology account; automatic ASR is not diarized or word-for-word audio verified. It does not disclose model versions, datasets, customer/product mapping, permissions, independent evaluation, or AI-attributed performance.
PredictNow AI / QTS Capital Management — Ernie Chan	Use GenAI to Manage Risk, Not Predict Return · Interactive Brokers — Machine Learning in Finance · Mutiny Fund episode 17 · timestamped queue ledger	Chan, introduced as the PredictNow AI CEO and QTS Capital Management founder, discusses regime change, data scarcity, risk management and portfolio optimization as more bounded applications than direct return prediction, and pretraining/fine-tuning on related time series as a proposal for limited data. The Interactive Brokers transcript adds a dated career and ML-in-finance publisher route; Mutiny Fund adds a separate QTS/Tail Reaper publisher route.	Practitioner/executive discussions, not QTS system disclosures. They do not establish a deployed pretraining pipeline, model inventory, permissions, or investment result.
Crescendo Partners / Ehsan Ehsani	Generative AI in Investment Management: Real-World Impact & Alpha Strategies · timestamped queue ledger	Ehsani is identified as an executive director at Crescendo Partners and describes GenAI use for activist-letter drafting and comparison, portfolio-thesis tracking, idea generation, and preliminary research at the top of the investment funnel. He keeps management judgment and long-term investment decisions with people and discusses buying/lightly customizing rapidly changing tools.	Dated practitioner account, not a formal Crescendo policy or model registry. It does not disclose model names, data rights, adoption measures, returns, agent permissions, or autonomous allocation.
AQR / Cliff Asness	Hoover Institution — Cliff Asness on Factor Investing and the History of Financial Economics · timestamped queue ledger	Asness describes Chicago/Fama research lineage, AQR’s historical factor foundations, increasing use of machine learning, named ML personnel, NLP for textual data and return forecasting, and continuing concern about intuition, overfitting, market impact, and transaction costs.	First-person executive account and historical lineage. It does not disclose AQR’s full current model inventory, data contracts, permissions, evaluation design, or AI-attributed performance.
Implied / Ying Hua	Fundamental Edge episode · public YouTube captions · Buzzsprout player · Odds on Open episode	Hua is identified as Implied founder/CEO and former Citadel/Balyasny PM. The 2026 captions add speaker-reported in-house live earnings-call transcription, raw-audio preservation, bespoke public-data scraping, human-rubric flagging, cloud Excel automation, scheduled earnings workflows, and a process-learning assistant (11:53–14:14; 15:42–17:10; 32:32–35:00; 39:24–49:46).	Current founder/product account plus historical former-employer context.
OneChronos / Kelly Littlepage	OneChronos CEO on Compute as a Trading Asset & Regulation Process · timestamped queue ledger	Littlepage describes an optimization-based institutional trading venue, machine learning as an enabler for combinatorial auctions, and a proposed compute futures/swaps market tied to compute, power, and energy.	Adjacent market-infrastructure executive account. It does not establish hedge-fund adoption, a live regulated product, model weights, customer permissions, or investment performance.
Fidelity Labs / Evan Schnidman	Alpha Intelligence — Scaling AI at Fidelity Labs · YouTube recording · capture ledger	The named Fidelity Labs leader describes problem-first product selection, rapid prototyping, coding-tool use by engineers, data quality as a condition for reliable outputs, smaller language models for focused financial contexts, knowledge graphs/ontologies, and AI-driven compliance/client-workflow examples.	Asset-manager incubator evidence rather than hedge-fund deployment evidence. The captions are automatic and the episode does not disclose model names, production permissions, adoption denominators, or investment outcomes.
SigTech / Bin Ren	Alpha Intelligence — SigTech’s Bin Ren on AI Agents for Investors · AI at Work: Finance — The Rethinking Work Show · SigTech media page · capture ledger	Ren describes MAGIC as a multi-agent investment co-pilot with specialist agents and a capital-markets knowledge workflow. The Rethinking Work publisher description adds a separate title-blind route about AI changing the design, testing, and execution of investment strategies, while identifying Ren’s former CIO role at Brevan Howard’s Systematic Investment Group.	Vendor/founder account and historical personnel context. No customer identity, agent roster, data rights, evaluation protocol, current Brevan deployment, or investment result is disclosed.
Former Two Sigma product leadership / Tharsis Souza	Alpha Intelligence — Unpacking DeepSeek and AI in Finance · public talks page · capture ledger	Souza discusses public-model training economics, supervised fine-tuning, reinforcement learning, inference/application-layer economics, and look-ahead bias when financial analysts use LLMs.	Historical personnel and methodology evidence. It does not establish current Two Sigma systems, a proprietary model, a financial backtest, or a trading deployment.
STAC / James Corcoran	Alpha Intelligence — How Wall Street Firms Are Actually Using AI · capture ledger	STAC’s Head of AI and Analytics discusses benchmarking models across hardware and cloud platforms, model-risk framing, provider-version drift, vector databases, data/model strategy, fine-tuning infrastructure, and model-as-a-service decisions.	Industry-observation and vendor-context evidence. No client identity, benchmark dataset, model score, or fund deployment is disclosed.
Alicia Vidler / agent-based market research	Alpha Intelligence — AI Agents in Markets · capture ledger	The episode discusses agent-based market simulation, document-grounded CLO/covenant analysis, numeric limitations, model-version concerns, human responsibility in regulated workflows, and a junior-analyst/support boundary for LLMs.	The episode’s Spotify description supplies historical personnel and AI-driven-fund context; that context requires independent cross-checking. No fund architecture, model, permission map, or investment result is disclosed.
Methodology / David Wright	Why generative AI still cannot trade	Separates predictive tree models and text-sentiment experiments; discusses quarterly retraining and temporal leakage.	Named-fund deployment or performance.
Acadian / Joseph Simonian	Uses and misuses of ML in finance	Short clip on research-program structure and statistical hygiene when moving from econometrics to ML.	Current Acadian model inventory.
Acadian / Scott Richardson	Credit Edge episode · publisher transcript mirror	Acadian’s first-party news page confirms Richardson’s Director of Systematic Credit role and a public discussion of systematic credit, alternative data, liquidity, transaction costs, leveraged loans, and CLOs.	Firm and podcast evidence does not disclose a complete credit-model inventory, data contract, trading authority, or independently measured result.
Systematica / Leda Braga	FEG Insight Bridge · official video · capture note	The dated video identifies Braga as Systematica’s CEO. At 44:00–48:24, the local caption capture records her public account of AI-assisted text extraction, agents constructing alphas from an investment thesis and data, agents implementing theses from papers and books, a domain-specific alpha language, cloud-scale computation, and detailed trade/transaction-cost records.	Executive self-report and official publisher video; no model inventory, agent count, training data, permissions, live authority, or performance attribution.
Systematica / Leda Braga	Goldman Sachs interview	The 2019 conference interview describes machine-learning alternatives for trend signals, blending five formulations, and dimensional reduction.	Dated executive self-report; no current model card, implementation, or performance record.
Systematica / Leda Braga	Women in Data Science Worldwide · Money Maze · SS&C event post	Title-blind publisher and event routes add Braga’s dated discussion of data science, trend-following, algorithm aversion, broad data types, and caution around autonomous decision-making.	Publisher metadata and event-organizer summary; no recovered full transcript, model inventory, permission map, or performance attribution.
Systematica / Leda Braga	CNBC Events transcript	CNBC’s public transcript attributes to Braga a separate data-research team that treats, interprets, and selects features from imagery, text, sound, unformatted text, and consumer data before alpha-making.	Event-organizer transcript of an executive appearance; supports a data-preparation organizational lane, not a disclosed model or production-authority map.
Systematica / Leda Braga	Portuguese-language Outliers episode · InfoMoney text route	Regional-language sources describe model-based decision-making, BlueTrend, and a human/machine boundary in which analysts can suggest directions for quantitative research.	Regional media and publisher metadata; audio and translation spot-checking remain open.
Systematica / operational resilience	SBAI Episode 20	The 2026 episode names Braga and Ben Dixon and discusses trade errors, near misses, transparency, disclosure, and no-blame operational culture.	Operational-risk evidence, not AI or model evidence.
Historical Brevan Howard / Bin Ren	Flirting with Models — text2quant · At the Forefront — Agentic AI · capture note	Ren’s former Brevan Systematic Investment Group route connects multi-asset backtesting, alternative data, LLMs as clients of financial data/services, and later multi-agent investment tooling.	Historical personnel and vendor media; no current Brevan deployment, model inventory, customer permissions, or performance result.
Brevan Howard / Sebastien Guglieta	The Quant Conference virtual-access index · capture note	The public 2019 London programme lists Guglieta as Brevan Howard’s Co-Head of AI Strategies Group for a keynote on AI-based macro strategy, human–machine collaboration, and causality. The recording is restricted to ticket holders, so the programme is a role/title lead rather than substantive presentation evidence.	Dated conference metadata only; no current role confirmation, model inventory, deployment, permissions, or performance.
Brevan Howard Centre / Sesh Karri	Imperial people page · research page · capture note	Imperial identifies Karri as a Research Fellow in Machine Learning working on ML for finance and tick-by-tick trading rules. His public page records research lineage through Francis Bach, Marc Deisenroth, Adrian Weller, and Vladimir Kolmogorov.	Academic-centre and researcher-lineage evidence; not evidence of Brevan fund employment, production models, permissions, or performance.

Historical Brevan and Caxton routes recovered without current-AI promotion

The title-blind publisher pass also recovered Bin Ren on AI at Work: Finance, which adds a separate AI-and-finance interview route and historical Brevan personnel context. It does not establish current Brevan deployment. A separate Bruce Kovner/Caxton episode is retained as historical founder media only. No performance figures from that episode are promoted, and it does not change Caxton’s current-AI classification. | Systematica / Data R&D and research platform | Culture page · Shanghai quant role · production systems profile | Public role surfaces mention a new Data R&D team, a proprietary high-level modeling language for forecasts, intraday-to-multi-month research horizons, robustness/overfitting, alternative datasets, and historical 24x6 production support. | Job and historical infrastructure evidence; no current organizational chart, model inventory, agent permissions, or performance attribution. | | GMO / Warren Chiang | Schwab Network archive · GMO role page | The platform archive lists a June 5, 2026 appearance and GMO identifies Chiang as a Systematic Equity portfolio manager. | The episode-level recording remains unresolved; this is personnel and systematic-investing evidence, not an AI-system disclosure. | | GMO / Tom Hancock | Masters in Business transcript | The title-blind transcript identifies Hancock’s historical machine-learning work at Siemens, his IBM/software background, his Harvard/MIT research-group route, Chris Darnell’s role in GMO’s quantitative-research origin, and Les Valiant as a PhD adviser. | Timestamped executive interview and historical lineage; no current GMO model, AI lab, or internal GenAI disclosure. | | GMO / Warren Chiang | Excess Returns episode · GMO webcast transcript | The 2025 podcast and GMO-hosted transcript expose systematic-value process vocabulary: financial-statement restatement, top-down/bottom-up construction, liquidity segmentation, risk constraints, and repeated research changes. | Named portfolio-manager media and first-party event material; no model architecture, data vendor, LLM use, or performance attribution. | | GMO / AI investment-theme materials | Hype vs. High Conviction · Rise of the Machines deck | GMO publicly frames AI through applications, LLMs, compute, suppliers, cash flows, capital spending, technological revolutions, bubbles, and capital cycles. | Investment thesis and conference education; does not establish internal AI/GenAI tooling. | | XAI Asset Management / Federico Fontana | AI won’t replace quant research | Harrington Starr identifies Fontana as CTO and describes a Quant Strats 2025 discussion of AI, quant research, trading infrastructure, smarter data, reinforcement learning, and systematic strategies. | Public interview scope and role only; no XAI model inventory, training corpus, permissions, or performance attribution. | | BlackRock Systematic / Jeffrey Rosenberg | FICC Focus | The episode identifies Rosenberg as a senior systematic fixed-income portfolio manager and discusses LLMs, systematic fixed income, rates, decorrelation, and a systematic ETF. | Podcast evidence does not establish BlackRock’s internal models, data rights, live authority, or AI-attributed result. | | Fidelity Systematic / Jessica Stauth | Alpha Exchange · transcript mirror | Stauth is identified as Fidelity’s CIO for Systematic Equities. The public discussion covers regime-aware strategies, overfitting guardrails, earnings-call text, ML tools, and LLM-driven sentiment extraction. | Public role and methodology discussion; no complete model inventory, data permissions, production authority, or independent performance attribution. | | WorldQuant / Third Point / Social Leverage / Matt Ober | Alternative Data Podcast · Odds on Open · David Spisak Show · ASR recovery note | The recovered January 2024 episode describes WorldQuant’s global data sourcing, ingestion and automation, and Ober’s Third Point remit to centralize data and make it usable in a traditional long-short process (02:21–05:56; 10:56–14:27). It also describes Initial Data Offering as a discovery channel for new datasets and names RavenPack, FinChat, Live Data Technologies, and DataVations as examples (28:41–35:05). | Historical personnel and speaker-reported workflow evidence; no current WorldQuant or Third Point architecture, proprietary data rights, model ownership, or investment result. | | Former Two Sigma quant / Omer Seider | Odds on Open · Spotify · capture note | The recovered audio describes a historical alpha-capture workflow for structuring expert opinions, then discusses digital analysts reading filings, news, transcripts, and factor data; context-overload and hallucination risks; proprietary-data differentiation; guardrails; and the distinction between generating analytics and exercising judgment. | Former-employee account; all Two Sigma references are historical and do not establish current Two Sigma systems, permissions, or performance. | | Evolution Exchange Singapore | AI in Hedge Funds panel · Apple mirror | Jiri Pik, Ernest Chan, and Jared Broad discuss decomposing investment work into bounded AI tasks, risk-scenario modeling, strategy optimization, data quality, compute costs, and explainability. | Regional practitioner/vendor panel; no single-firm deployment, customer identity, model version, permission, or performance is disclosed. | | Peterson Capital Management / Matthew Peterson | This Week in Intelligent Investing | The managing partner discusses an internally developed AI research platform alongside a value-oriented investment and options process. | Adjacent investment-manager case; no platform architecture, customer permissions, evaluation results, or AI-attributed returns. | | Arrowpoint Investment Partners / Jonathan Xiong | Macro Hive episode · Libsyn player · timestamped capture note | The Aug. 28, 2025 interview identifies Xiong as Arrowpoint founder and CEO/CIO and discusses AI agents approaching analyst-like research capability, the difficulty of connecting agents to proprietary/broker/expert-network/transcript data, possible document and legal retrieval, and organizational adoption friction (00:15:51–00:19:58). | Dated speaker account and recovered local ASR; no named Arrowpoint model, agent, vendor, training corpus, data rights, production permission, headcount change, autonomous trading authority, or AI-attributed result is disclosed. | | L&G Asset Management / Systematic Solutions | First-party team podcast | L&G names its Head of Systematic Solutions and quant/factor and index specialists and discusses index engineering, AI, megatrends, and specialist data. | Asset-manager first-party evidence; no model weights, training data, permissions, or independent outcome. | | Bloomberg Volatility Forum Singapore | FICC Focus recording | The conference recording names Capula and Dymon Asia portfolio managers alongside Bloomberg, Optiver, and Quadriga participants in volatility and derivatives sessions. | Conference and personnel context only; AI or hedge-fund workflow evidence remains unresolved. | | Horizon Investments / Mike Dickson | Money Path episode · YouTube captions · capture note | Horizon identifies Dickson as Head of Research and Quantitative Strategies. Recovered YouTube captions add timestamps for the episode’s AI-outlook discussion: he says hyperscaler AI demand raises the earnings bar and that productivity effects should appear in margins and output (00:05:08–00:05:37). | First-party page plus platform-generated captions; direct audio returned HTTP 403. This is investment-outlook commentary, not evidence of a Horizon production model, data permissions, or independent AI attribution. | | Rayliant / Research Affiliates / Jason Hsu | Excess Returns episode · YouTube | Hsu is presented as Rayliant CIO and Research Affiliates co-founder; the episode discusses China, factor investing, machine learning in factor construction, factor decay, and AI’s impact on investing. | Public systematic-investing discussion; no complete Rayliant model inventory, training corpus, data permissions, or independent results. | | Minotaur Capital (Australia) | Armina Rosenberg / Thomas Rice · NAB Morning Call · Minotaur’s first-party release · capture note | The NAB interview adds a recovered, timestamped practitioner route: Rosenberg describes a multilingual news funnel, AI-generated snapshots and initiations, continuous stock agents, thesis-validation tests, portfolio construction retained by the two founders, and an approximately 20-model API layer with thousands of daily calls. The episode says 173 sources; later Minotaur material says 174, so the discrepancy is retained. | Publisher-hosted interview and firm/speaker accounts; model roster, versions, data rights, error rates, permission mechanics, staffing audit, and independent return attribution remain undisclosed. | | Plato Investment Management (Australia) | Marcus Howes / Livewire plus Plato team page | Public description of a text-first earnings-call Q&A-evasion review trigger using topicality, answer/question length, and future-tense features; Plato identifies Howes with ML/NLP/LLM experience. | Semantic triage, not acoustic lie detection; no public permission map, live trade authority, independent audit, or validated lead time. | | Plato Investment Management — automation and research personnel | Plato team page | Wilson Thong is publicly described as overseeing automation across production, compliance, and reporting and as creator of PRISM; Chanel Stuart-Findlay’s profile records prior ML/NLP investment-process research. | Personnel pages describe scope, not current model ownership, deployment stage, or capital authority. | | PGIM Quantitative Solutions / Stacie Mintz | Flirting with Models — S7E33 · PGIM biography · capture note | The publisher and PGIM identify Mintz as Managing Director and Head of Quantitative Equity. At 38:33–45:54, the recovered audio records a staged LLM research process, LLM/small-language-model use for non-financial quality measures, text normalization, look-ahead and memorization controls, model-selection and out-of-sample checks, multi-model comparison, repeat-prompt stability, and linkage back to company fundamentals. | Dated practitioner account plus first-party role corroboration; no PGIM model inventory, training data, permissions, or independently attributed performance. | | Campbell / Joseph Kelly | UBP: AI, macro events, and diversified returns | UBP identifies Kelly as Campbell’s Managing Partner and describes Campbell’s progression from AI-assisted research and code completion toward agentic coding tools, including Claude Code. | Named executive account on a UBP page; no model inventory, data rights, permissions, evaluation results, or return attribution. | | Money Maze / Andrew Veglio, Kristian West, Lucia Soares | AI and the Investment Edge · YouTube recording · capture note | A dated compilation identifies the three speakers and records West’s discussion of data foundations, research consumption, approximately 7,000 broker reports per day, an agentic framework keyed to portfolio holdings, and agent context/style; Soares discusses employee adoption and Veglio advances a replacement thesis. | Publisher metadata plus automatic YouTube captions; speaker claims are not independent audits and do not establish model identity, permissions, production status, or investment outcomes. | | Capital Fund Management / Philip Seager | AIMA: Decoding quant · archived audio/ASR ledger | The recovered audio identifies Seager as CFM’s Head of Portfolio Strategy and records firm-reported discussion of model-based research, robustness testing, more than 100 research staff, alternative data, integrated ML tools, a shared research/data platform, cloud compute, and common internal tooling. | Dated executive account and local ASR; no CFM model registry, data contracts, current headcount audit, permissions, or performance attribution. | | Capital Fund Management / AI-native quant researcher | CFM careers page | A public role describes a newly formed quant-research team building AI-based alpha models, with dataset construction, training recipes, foundation-model architectures, multi-GPU training, evaluation, monitoring, and ML-platform collaboration. | Hiring signal only; no evidence of hiring completion, model ownership, production deployment, permissions, or performance. | | Capital Fund Management / ML Platform | ML Platform Engineer role · CFM approach | CFM describes a platform remit spanning large-scale market data, data→training→evaluation→deployment, MLOps, reliability, monitoring, reproducibility, quality, auditability, and cloud/ML integration. | First-party architecture and hiring language; no independently audited platform, adoption, or model outcome. | | Capital Fund Management / alternative data and academic partnership | Data Scientist — Alternative · CFM–Columbia announcement | CFM’s role description names alternative-data feature extraction, time-series ML, LLM-assisted coding, and agentic AI. The historical Columbia announcement describes alternative-data and economic-forecasting research. | Public hiring and historical partnership signals; no current dataset, researcher roster, deployment status, or independent model evaluation. | | CFM ML Lab / Eric Vanden-Eijnden, Anastasia Borovykh, and Giulio Biroli | CFM public ML Lab post · CFM academic partnerships | CFM names a new ML Lab, its named leadership, and collaboration with the CFM–ENS Data Science chair. The stated remit covers ML/LLM mechanisms and limitations, financial applicability, and collaboration with CFM researchers. | Public company/social and academic-partnership evidence; no complete lab roster, model cards, production permissions, or performance results. | | CFM ML Lab / generative-models seminar and postdoctoral route | CFM seminar post · CFM-ENS postdoctoral announcement · capture note | CFM identifies Eric Vanden-Eijnden as Head of Machine Learning and describes a New York seminar on generative models for quantitative finance, signals, and models. A separate post advertises one or two two-year postdocs on diffusion models and generative-AI theory, supervised by Giulio Biroli through the CFM-ENS/ML Lab route. | First-party/social and academic-recruiting evidence. No seminar recording, model specification, filled-hire confirmation, live permission map, or performance attribution. | | CFM / Christian Dery — GenAI strategy | CFM: Investing After the AI Honeymoon · CFM: systematic global macro | CFM states that it is investing in generative AI and names sentiment/context extraction, classification, tokenized prediction problems, risk management, automated research, coding, and LLM-proposed network/classification structures as use-case areas. | First-party strategy commentary and sponsored insight; no model inventory, benchmark, production authorization, or return attribution. | | CFM / Probabl investment | Probabl seed announcement | Probabl says CFM co-led its seed financing alongside Serena. This adds an external open-source ML infrastructure investment signal to CFM’s public AI ecosystem. | Financing participation does not establish a commercial partnership, software adoption, data sharing, or model integration at CFM. | | CFM ML Lab / academic bridge | P&I interview · CFM strategy interview · CFM approach · capture note | CFM’s first-party material describes ML Lab members embedded with research teams, an academic-bridging remit, multimodal extraction from text/video/audio into predictors, 12+ PB of financial and alternative data, and pre-deployment testing with board-level override authority. | First-party organizational and strategy evidence; no full lab roster, model weights, training corpus, permissions, evaluation results, or AI-attributed returns. | | Morgan Stanley QIS / Stephan Kessler | HFR: AI and Quant — Revolutionizing QIS | The first-party transcript covers AI-assisted signal generation, sentiment analytics, alpha decay, factor characterization of hedge-fund returns, and portfolio construction. | QIS methodology and industry context; no Morgan Stanley model inventory, client deployment, data permissions, or investment outcome. | | Russell Investments / Kate El-Hillow | The Private Capital Podcast | The publisher identifies El-Hillow as President and CIO and describes neural networks applied to proprietary investment data for manager research, simulations, and information filtering. | Publisher description and executive interview; no model specification, dataset construction, validation design, permissions, or investment attribution. | | Galaxy Digital / Will Owens and Zack Pokorny | Institutional Edge: Pricing Uncertainty · linked Galaxy research | The episode identifies Galaxy research and data personnel and discusses prediction or signal markets, information aggregation, and AI agents as liquidity or market-making infrastructure. | Adjacent market-design evidence; no hedge-fund production agent, model details, customer permissions, or investment results. | | Historical Flow Traders / Tower Research personnel / Annanay Kapila | Flirting with Models · Rareliquid transcript · Basis Points · SWE Accelerator / Quant Historian · QFEX biography · Basis Points capture note · Quant Historian capture note | Multiple public surfaces identify Kapila as a former Flow Traders and Tower Research quantitative trader and discuss HFT, perpetual-futures design, market structure, risk management, and model lifecycle. The August 19, 2026 Basis Points recording adds a title-blind account of short-horizon ML features, fair-price/data checks, and AI-assisted coding and support-log workflows. The March 15, 2026 SWE Accelerator episode adds a full timestamped transcript and recovered public MP3 for a longer discussion of HFT/market-making concepts, quant-industry history, model-risk anecdotes, and automation. | Historical personnel and market-structure evidence plus current-company self-report; neither episode establishes current Flow/Tower systems, QFEX model ownership, permissions, independently measured productivity or loss figures, or performance. The local ASR used for the recovered MP3 is non-diarized and navigation-only; exact quotations require audio checking. | | Quant workflow methodology / Mattia Spreafico | The Blushing Quants · Apple mirror · audio-backed capture | A Switzerland-based quant discusses LLMs as tools for research, coding, and iteration, while emphasizing data access, monitoring, controls, deployment constraints, and the speed-versus-review trade-off. The publisher MP3 now has a local timestamped ASR pass, including discussion of human review and production gates. | Methodology and biography evidence only; automatic ASR is not diarized or word-for-word audio verified. No current employer, firm-specific deployment, model inventory, data rights, or investment outcome is established. | | MarginLens / historical bank quant context | Talking Tuesdays with Fancy Quant · Amazon Music mirror | The publisher describes a founder identified only as Caitlyn, with prior JPMorgan and Bank of America quantitative-engineering experience, building an AI-powered margin-intelligence platform and discussing documentation, architecture, and explainability. | Adjacent financial-infrastructure evidence; surname, model inventory, customer permissions, validation results, and hedge-fund deployment are not established. | | AQR / Peter Hecht | Flirting with Models · AQR portable-alpha paper | AQR’s co-head of North America Portfolio Solutions discusses portable alpha, funding, tracking error, tail correlation, sizing, rebalancing, and liquid wrappers; AQR separately published a dated portable-alpha paper with Hecht. | Portfolio-solutions evidence; no AQR AI model inventory, data permissions, trading authority, or independent performance attribution. | | Caladan / John Gu | Flirting with Models · Caladan company post | Caladan’s CEO discusses crypto market-making, liquidity bootstrapping, inventory and portfolio risk, adverse selection, arbitrage, structured products, and treasury solutions. | Crypto market-making and infrastructure evidence; no Caladan AI model inventory, data rights, execution permissions, or performance. | | Macquarie QIS / Faheem Osman | Flirting with Models · Podscan transcript · LinkedIn | Osman is identified as Macquarie’s Global Head of QIS Structuring at recording time; the episode covers commodity QIS, curve carry, volatility premia, factor taxonomy, and overfitting. | Bank-QIS evidence, time-scoped to the recording; no Macquarie model registry, hedge-fund customer system, data permission, or independent result. | | Numerai / Richard Craib | Flirting with Models | Numerai’s founder discusses crowdsourced machine learning, data obfuscation, residual correlation, adversarial behavior, stake-weighted meta-models, orthogonal alpha, and risk-management changes. | Founder discussion complements newer Numerai model disclosures; it does not independently establish current model size, training corpus, permissions, or performance. | | AlphaNova / Marc Nunes | The Sophron Network episode · timestamped local transcript · capture note | The July 28, 2026 episode identifies Nunes as AlphaNova’s co-founder and CEO. He describes leakage checks, overfit screening, submission limits, a live observation period, decorrelation, a neural-net prediction layer, manual Claude-assisted signal review, an intended contest-to-trading pipeline, and an agent-swarm workflow for research ideas. | Named practitioner and company-methodology evidence. The local ASR is not diarized; no model weights, training corpus, permissions, current signal roster, independently audited performance, or current deployment date is established. | | Bloomberg Research Data / Angana Jacob | Flirting with Models · Bloomberg quant-data and AI panel | Bloomberg’s Global Head of Research Data discusses data-set lifecycle, build-versus-buy, geographic exposure data, quant pipelines, and raw versus value-added modeling; Bloomberg separately places her in AI-ready investment-data panels. | Vendor and conference evidence; no client-specific model access, hedge-fund deployment, or performance. | | Takahē Capital / Moritz Heiden and Moritz Seibert | Flirting with Models · Takahē team · University of Augsburg profile | The quant manager discusses trend and spread strategies in thin futures markets, capacity, alternative data, and the decision to keep a global-markets fund small. First-party and university pages establish Heiden’s quantitative-research role and academic lineage. | Small-manager and academic-lineage evidence; no AI/GenAI deployment, model registry, trading permission, or independent performance attribution. | | Man Numeric / Jayendran Rajamony | Flirting with Models · Man Group personnel page | Rajamony is identified as Man Numeric’s Director of Alternatives and discusses quant-equity evolution, alternative data, covariance noise, operational risk, and systematic/discretionary boundaries. | Named systematic-process evidence; no Man Numeric AI model inventory, data contracts, permissions, or AI-attributed result. | | Simplify / Roxton McNeal and Siddharth Sethi | Flirting with Models · Simplify fund document | Simplify’s QIS leaders discuss multi-strategy QIS portfolio construction, bank-offering evaluation, dynamic allocation, infrastructure, and risk management; the fund document independently lists the named portfolio managers. | QIS and personnel evidence only; no AI system, model inventory, data rights, or independently measured outcome. |

Newly recovered title-blind audio and video — August 18, 2026

Firm / person	Recovered source	Publicly observable signal	Boundary
Two Sigma / Ben Wellington	Flirting with Models — S7E32 · Two Sigma first-party follow-up · capture note	Wellington describes a feature layer between raw data and forecasting, a shared platform that lets teams reuse features and ML tools, LLM-generated text as a new data surface, hypothetical computer-vision exploration of earnings-call video, and an AI strategy centered on amplifying researcher originality and orthogonality.	Named practitioner account and first-party summary; no model registry, training corpus, permissions, live authority, or performance attribution.
PGIM Quantitative Solutions / Stacie Mintz	Flirting with Models — S7E33 · capture note	Mintz describes a staged, research-first LLM workflow: start with a concept or data set, use the simplest suitable tool, test shorter historical windows, compare multiple models and repeated prompts, and connect extracted insights back to company fundamentals.	Dated executive/practitioner account plus PGIM role pages; no complete model inventory, data permissions, production authorization, or independently attributed result.
Systematica / Leda Braga	FEG Insight Bridge · official YouTube recording · capture note	Braga publicly describes agents constructing alphas from investment theses and data, agents implementing ideas from papers and books, a domain-specific alpha language, cloud computation, and hundreds of thousands of computed alpha constructs.	Named CEO interview and official video captions; no agent count, model family, training data, permission map, live trading authority, or performance attribution.

Newly promoted practitioner and operating-model sources — August 18, 2026

Firm / person	Recovered source	Publicly observable signal	Boundary
Balyasny / Giuseppe “Gappy” Paleologo	ACE Talks episode · Balyasny profile · capture note	The dated interview and first-party profile establish Paleologo’s Global Head of Quantitative Research role and research lineage. He describes AI as accelerating problem formulation and iteration, while placing a boundary around repetitive work and the originality involved in discretionary interpretation.	Named practitioner testimony and personnel evidence; no Balyasny model inventory, training corpus, agent permission, production deployment, or investment result.
Man Group / Greg Bond	Hedge Fund Huddle · capture note	Man Group’s CIO names AlphaGPT as an internal technology spanning data onboarding, hypothesis testing, backtesting, and a possible investment-process layer. He also describes selecting different language models for coding, hypothesis generation, and writing, with adoption and value as practical tests.	Executive account through an LSEG publisher transcript; no model versions, data contracts, production coverage, error rates, live permissions, or AI-attributed returns.
Point72 / Jaimi Goodfriend	Hedge Fund Huddle — Point72 Academy · capture note	The Academy director describes an in-house apprenticeship and an updating curriculum covering which AI tools to make available, how to train analysts to use them, and how processes should change. The episode also identifies a separate Cubist Academy for systematic investing.	Training-leader testimony; no Point72 or Cubist model inventory, tool-permission matrix, live trading authority, customer data boundary, or investment outcome.
Versor / StarMine	LSEG Hedge Fund Huddle — Quant Questions · capture note	LSEG’s transcript identifies Versor founding partner Nirav Shah and StarMine senior quant Tarun Sanghi. Sanghi describes clear-box stock-selection and risk models, a historical text-mining credit model, and a merger-and-acquisition target model combining fundamental and alternative data with BERT-based text processing. Shah gives user-review sentiment as an alternative-data example and describes AI across alpha research, risk, asset allocation, and portfolio construction.	First-party publisher transcript and named practitioner account; no current model weights, training corpus, data rights, customer permissions, evaluation splits, or AI-attributed performance. Historical StarMine product language is not assigned to current Versor architecture.
LSEG / Rational AI / Capital One / Principals Media	Hedge Fund Huddle — Separating best practice from hype · capture note	The LSEG event transcript names a text-analytics product leader, a former Bloomberg news-product leader, Rational AI’s founder with former Two Sigma context, and Capital One’s machine-learning leader with former quant-trader and Bloomberg-engineer context. The discussion frames finance LLM adoption around data privacy, model accuracy, regulation, and a publisher-reported 55-firm survey conducted in November 2023.	Panel and personnel evidence; the underlying survey instrument and any employer-specific system details were not recovered. Do not treat panelist backgrounds as evidence of a current or former employer’s model inventory or adoption rate.
LSEG / Epistemic AI	Hedge Fund Huddle — What hedge funds can learn from healthcare · capture note	The first-party transcript identifies Epistemic AI co-founder Stefano Pacifico and describes AI/NLP knowledge discovery across fragmented biomedical sources. The discussion covers why language models are not databases, incomplete retrieval, hallucination, sequence construction, and combining language models with other systems.	Adjacent biomedical methodology and failure-mode evidence; no hedge-fund deployment, clinical-trial approval predictor, customer identity, technical benchmark, or financial use is established.

Balyasny applied-AI media recovery — August 19, 2026

The title-blind pass recovered a distinct Balyasny media cluster that was not represented by the earlier podcast seed queue. The Bloomberg Tech Disruptors episode is a dated publisher page with Charlie Flanagan discussing generative AI, investment workflows, developer productivity, hallucinations, GPU supply, and automation. The The TRADE interview adds a separate practitioner account of a centralized AI team, quick-answer and deep-research workflows, user education, and cross-team reuse of lessons.

A further public surface is the March 6, 2026 OpenAI customer story, which describes a centralized Applied AI group, internal model evaluation, task-specific model selection, agent workflows, traceable reasoning, real-time feedback, scoped data/tool access, and multimodal and reinforcement-fine-tuning directions. The Balyasny announcement confirms that the firm endorsed and linked the story. These are attributed vendor and firm statements; they do not provide independent benchmark code, model weights, data entitlements, permission maps, or investment attribution.

A later check of Balyasny’s official YouTube channel recovered automatic captions from several short, title-blind clips. How My Team Develops Investment Theses describes combining market-structure understanding with data-science or machine-learning methods, testing hypotheses empirically, and iterating through validation, refinement, or rejection toward operational deployment (00:00–00:50). Gappy Paleologo on How QR Partners with Investment Teams describes quantitative researchers supporting portfolio managers with day-to-day portfolio management, performance understanding, and new data (00:03–00:32). A Troy Gao clip adds recruiting language about historical patterns, large data sets, probability, and the interaction of systems, infrastructure, and process (00:05–01:17). These are firm-controlled short-video statements, with automatic captions used as locators; they do not disclose model identity, training data, permission boundaries, production coverage, or performance. The capture note records the full checked set and exclusions.

The April 2026 Balyasny AI Hackathon page also embeds a firm-controlled recap video. Its manual caption track describes concurrent events in Singapore, London, and New York with investment and technology staff working together (00:05–00:15), agents being used to create workflows with examples involving investment-team previews, technology vulnerability searches, and proactive data surfacing (00:40–00:55), and a speaker statement that earlier AI efforts began with a hackathon around the first model (00:56–01:08). This adds a concrete cross-region, cross-function operating signal to the article’s existing hackathon record. It does not name participants or winning projects, identify models or datasets, establish production rollout, or show individual ownership or measured investment impact. See the updated capture note.

The Artefact/Arthur AI session adds a vendor-partnership surface around retrieval, hallucination controls, and investment-research workflows. Its public LinkedIn event post also exposes a publisher transcript covering the centralized team, the retrieval corpus, hallucination controls, productivity use cases, and model selection/performance monitoring. The post’s short link now resolves to a YouTube replay with timestamped English automatic captions. Those captions add navigation anchors for the six-million-document retrieval description, claim-level hallucination checks, management-meeting question generation, recurring model monitoring, and finance-text embeddings (03:35–06:17; 10:00–15:03). This remains vendor/customer-session evidence and is not an independently audited technical record. See the updated capture note.

The full dated source matrix, personnel graph, and unresolved capture list are in the Balyasny media expansion note.

A title-blind Capital Allocators interview with Dmitry Balyasny, uploaded October 2, 2023, supplies a broader platform baseline. Balyasny describes entering new strategies through research, a staged operating blueprint, guardrails, specialist hiring, research and technology stacks, risk management, and portfolio construction (15:54–18:45). He describes cross-strategy sharing around data monetization, technology, and portfolio construction (25:53–27:40; 39:04–40:20) and says large technology, data, and legal teams work across strategies (70:01–70:48). The recovered caption transcript contains no explicit AI, GenAI, language-model, or agent discussion. That is a negative-control result for this dated interview, not evidence about the rest of the firm’s media record. See the capture note.

A title-blind Meb Faber interview with Vinesh Jha adds a former-PDT and alternative-data-provider route. The publisher identifies Jha as ExtractAlpha’s founder and says he was most recently an Executive Director at PDT; the interview describes entity mapping, point-in-time and survivorship-bias controls, hypothesis testing, and research on earnings-call NLP, Chinese news sentiment, patent/visa hiring data, and crowdsourced earnings estimates (12:50–26:41; 43:50–49:03; 52:14–64:45). ExtractAlpha’s current Solutions catalog lists US and Japanese earnings-call AI models, Japan news/POS data, an innovation model using patents and visas, digital-demand data, and Estimize. Its team page names Yunan Liu, Qayyum Rajan, and Alan Kwan and records their academic or prior-firm affiliations. These are dated practitioner and first-party provider signals, not evidence of current PDT usage, a specific hedge-fund customer, model weights, data rights, or investment performance. See the capture note.

The same speaker-name expansion recovered Harold de Boer of Transtrend on The Derivative and a separate Top Traders Unplugged interview that explicitly discusses whether generic ML/AI language corresponds to robust trend-following practice. Transtrend’s first-party role page anchors the speaker-to-firm relationship. These are methodology and boundary signals, not evidence of a current Transtrend model inventory or deployment.

Additional title-blind systematic sources — August 18, 2026

Firm / person	Recovered source	Publicly observable signal	Boundary
Aspect Capital / Martin Lueck	J.P. Morgan Trading Insights · capture note	Aspect’s co-founder and Research Director describes a model-first research process, constrained data experiments, interpretability, and use of ML across signal generation, volatility forecasting, portfolio construction, and execution research. LLMs are framed as useful research tools whose outputs require explanation and controls.	First-party publisher transcript and named practitioner testimony; no complete Aspect model registry, training corpus, production coverage, permissions, or performance attribution.
Aspect Capital / Razvan Remsing	Aspect first-party announcement · ReSolve recording · publisher transcript PDF · capture note	The 2021 Director of Investment Solutions interview has recovered speaker-labeled timecodes. Remsing discusses hypothesis-first models, flow and sentiment data, forward-looking information, conditional models, risk forecasting, and model diversity across market environments.	Dated practitioner account and publisher transcript; no current Aspect model inventory, GenAI deployment, data rights, production permissions, or performance attribution.
Unlimited Funds / Bob Elliott	Behind the Ticker · YouTube recording · capture note	The publisher connects Elliott’s Bridgewater background with Unlimited’s hedge-fund-replication product and describes a proprietary Bayesian ML approach that treats inferred positioning as path-dependent rather than relying only on long rolling regressions.	Company-reported product architecture and historical personnel context; the page’s performance discussion is hypothetical/back-tested and does not establish realized returns, customer-specific deployment, or Bridgewater’s current systems.
QuantEdge Capital / Suhaimi Zainul-Abidin and Leonardo Jong	Odds on Open · The Edge Singapore profile · capture note	The regional source describes a systematic manager’s stated separation between GenAI-assisted research, data cleaning, execution, and programming on one side and a deterministic, interpretable core portfolio process on the other. It also adds Singapore/Asia personnel and organizational context.	Podcast metadata and a profile/advertorial surface; no model inventory, vendor list, evaluation protocol, employee permissions, or independently audited performance is established.
Vltava Fund / Daniel Gladiš	Strand Global Macro interview · YouTube recording · Spotify episode · capture note	The publisher identifies Gladiš as Vltava Fund CIO and the recovered automatic captions provide timecoded discussion of AI-assisted company research, information processing, intellectual conformity, and the human boundary around reliability, valuation, conviction, and responsibility.	Publisher metadata and automatic-caption evidence; no speaker-verified transcript, model inventory, data source, AI workflow beyond the stated scope, permission map, production evidence, or performance attribution.
Strand Global Macro archive / 16 additional manager and economist routes	Archive enumeration · full capture note · caption recovery	The archive exposes separate YouTube and Spotify routes for Quantica, Quantedge, Lombard Odier, Aberdeen Investments, Maat Investment Group, Crossbridge Capital, AXA Group, Iguana Investments, Frazis Capital Partners, Eurizon SLJ, Syz Group, JM Finn, Sionna Investment Management, Kernow Asset Management, Umbra Capital, and Oldfield Partners. YouTube automatic-caption layers have now been recovered for the full 17-episode set. The final five add title-blind manager-process evidence: Crossbridge and Sionna discuss AI as an investment/valuation theme; Kernow contrasts quant short-horizon trading with its own long/short process; Oldfield describes AI-related analysis as largely qualitative; and Umbra’s CIO reports a trial of Anthropic Claude for consolidating internally aggregated data used in asset-allocation work and client/prospect literature (40:20–41:05).	Publisher metadata and automatic-caption navigation; the Umbra statement is speaker-reported and dated, not an independent deployment audit. None of these recordings establishes a complete model inventory, data rights, evaluation, permissions, production authority, or performance attribution.

| Atreides Management / Gavin Baker public-media cluster | Capital Allocators · a16z Runtime · Colossus episode 451 · Colossus episode 473 · Colossus episode 485 · Generating Alpha episode 56 · Runtime recording · Colossus 451 recording · May recording · August recording · Capital Allocators public enclosure · capture note | Six dated publisher routes identify Baker as Atreides managing partner/CIO and expose public discussion of crossover investing, AI infrastructure, model economics, model routing, and a show-notes timestamp labeled “How Gavin Uses AI in Atreides.” Recovered Runtime captions add navigation for model economics and outcome-linked software discussion (23:07–28:07); episode 451 captions add navigation for compute economics and agent-mediated software workflows (58:29–59:18; 1:12:16–1:15:27). The May recording’s captions add navigation for proprietary-data, model-economics, distillation, and cybersecurity discussion (54:58–59:18). The August episode’s show notes add model routing (00:26:38), continual learning (00:23:55), and token-spend (00:30:51) markers; the recovered captions also identify a separate Rogo/Felix advertisement (12:06–12:33) that is excluded from firm evidence. Local ASR of the Capital Allocators enclosure adds navigation to historical factor-risk training (00:45:07–00:46:01), Baker’s public/private AI-investing lens (00:53:59–00:54:48), public research and hiring signals (01:02:56–01:03:35), and a general view on AI narrowing execution gaps (01:12:01–01:12:31). None of these recordings verifies a specific Atreides workflow. | Publisher metadata, show notes, automatic captions, and non-diarized local ASR; the captions/ASR are not treated as verbatim transcript evidence. No model inventory, data-rights record, budget, permission map, production endpoint, or performance attribution is inferred. |

Institutional and regional title-blind expansion — August 18, 2026

Source	What the public page establishes	Boundary and next step
PGIM Quantitative Solutions / George Patterson	Top Traders Unplugged identifies Patterson as CIO of PGIM Quant Solutions and timestamps discussion of regime detection, machine learning, portfolio construction, and language models in investment research.	Publisher metadata and topic index, not a model inventory or deployment record. Recover the audio/transcript and cross-check PGIM first-party pages.
CFA Institute / Francesco A. Fabozzi · Brightcove captions	The CFA Institute Research and Policy Center’s dated video and public WebVTT captions cover LLMs in quantitative and discretionary research, unstructured-data signals, predictive factors, portfolio construction, hallucination risk, evaluation datasets, coding agents, and governance.	Institutional methodology context, not evidence that any specific fund uses the described methods. Captions establish what was published, not a firm deployment or performance result.
CPP Investments / Investment Magazine teaser · official symposium agenda	The agenda schedules the CPP session for 15 October 2026 and names Jon Webster and chair Lachlan Maddock. The teaser frames experiments, disappointments, and technology-stack design as planned discussion topics.	Upcoming event, not a missing recording. Planned topics do not establish what CPP has deployed. Track the session for a later recording or transcript.
Block Central Chinese-language episode · Bilibili video	Bilibili DASH recovery and local Chinese ASR now provide timestamped navigation. Charles Shang identifies himself as BCY Labs CTO and describes Transparent and Verifiable AI, factor decay, text/sentiment factorization, agent/harness coordination, and a proposed multi-agent world-simulation sandbox (01:45–02:19; 08:03–08:39; 26:23–27:17; 37:46–39:17). He characterizes current industry use as more often AI-assisted analysis and factor extraction than autonomous capital allocation (21:43–22:45).	Automatic Chinese ASR remains a navigation aid pending manual audio reconciliation. The discussion does not establish a third-party fund’s deployed system, model inventory, data rights, permissions, or results.
Robeco Quant Podcast — German Episode 5 · YouTube captions	Robeco’s firm-controlled episode is dated 24 June 2025 and identifies Dr. Matthias Hanauer in the Quant Research team. The German automatic captions describe nonlinearities and interactions, a roughly five-year infrastructure build, domain expertise and human direction in low-signal-to-noise finance, and NLP/context-aware language models for sentiment, risk, and investment-theme discovery.	Firm-hosted practitioner account and automatic-caption evidence; no model inventory, training corpus, permissions, production stage, or performance attribution.
Heidrick & Struggles / Arnaud de Servigny · recovered audio	The dated interview discusses controlled LLM use, code review and training, agent supervision, human judgment, and possible future technical/data-driven CIO roles.	Interview perspective, not a disclosure by a named hedge fund. Local ASR time ranges remain research aids pending manual speaker verification.
FondsGedanken / LOYS and Private Alpha Germany · RSS · recovered MP3	The 29 April 2026 episode names Dr. Julian Kauffeldt, Christoph Gum, and Paul Barthels. Local German ASR surfaces discussion of AI-assisted research, curated data inputs, earnings-call triage, and research augmentation at discrete time ranges.	ASR-derived statements require manual speaker verification and are not promoted as independently verified deployment or performance evidence.

New cross-surface sources recovered by guest, role, vendor, and conference searches

These sources were absent from the original seed queue because their titles or distribution surfaces did not use the expected hedge-fund/quant/AI vocabulary. They are included as evidence-bearing records with source-specific limits.

Firm / people	Canonical surface	What it adds	Boundary
BlackRock Systematic / Raffaele Savi	Exchanges episode	The 12 August 2025 episode identifies Savi as Global Head of BlackRock Systematic and describes quantitative investing, AI, and market-volatility discussion.	Publisher metadata and named role; no transcript or model/deployment evidence.
Fundamental Edge / Portrait Analytics	Other People’s Money episode	Brett Caughran and David Plon discuss adding AI to investment processes, thesis monitoring, adoption, specialty tools, custom versus off-the-shelf systems, prompting, and failure modes.	Vendor/practitioner discussion; no client-specific deployment or performance evidence.
Unnamed credit hedge fund / Sachin Kullkarni	Hedgineer Episode 4 · ASR capture note	Apple’s description names Kullkarni as Head of Data Science at an unnamed multibillion-dollar credit hedge fund and notes prior Point72 experience. Recovered audio adds a stack consisting of Jupyter, AWS/S3, Snowflake, Tableau/QuickSight, Airflow, and Materialize, and discusses LLM-assisted coding, summarization, unstructured-to-structured mapping, and reference-data glue (00:31–00:48; 17:19–22:34).	The employer is intentionally unresolved; the tools and workflow are speaker-reported and not evidence of a named fund’s current deployment, model inventory, data rights, or performance.
Ridgewood Investments / Sam Namiri	Planet MicroCap episode · capture note	The June 3, 2026 episode identifies Namiri as Partner and Portfolio Manager and describes AI layered onto valuation/company/sector screening to identify potentially temporary problems for management-meeting verification (13:29–15:11). He also recounts a Gemini web-traffic hallucination compared with paid sources (09:18–10:01).	Speaker-reported workflow based on automatic ASR; no model version, data license, evaluation, autonomous execution, or portfolio attribution is disclosed.
Wellington / Mark Sullivan and Roberto Isch	Capital Allocators Episode 427 · recovered Libsyn audio · capture note	The recording names Sullivan as Head of Wellington’s Hedge Fund Group and Isch as a risk and portfolio manager. It describes the evolution of Wellington’s hedge-fund platform, manager research, portfolio construction, quantitative and qualitative underwriting, and layered risk oversight.	Public practitioner testimony and local ASR. It does not establish AI use, model ownership, permissions, or performance; the described process is not treated as an AI disclosure.
Quant Strats EU 2025	Official brochure	The 14–15 October 2025 London programme exposes sessions on reinforcement-learning agents, LLMs and latent signals, unstructured alpha, alternative data, infrastructure, and portfolio risk, plus named speakers across Millennium, Citadel, Man Group, BlackRock, Robeco, LGT, and others.	Conference brochure evidence; no attendance, recording, live deployment, or performance inference. Promotional brochure language is excluded.
LGT Capital Partners / Roger Hilty	LGT event announcement	A first-party announcement places Hilty on the October 2025 Bloomberg Hedge Fund Forum programme in an AI/ML strategy context.	Event announcement only; no recording or substantive disclosure captured.
金声玉亮 / Lei Zong	Chinese-language Apple episode	Chinese-language metadata identifies Lei Zong as founder and exposes a discussion of AI quantitative investing and a former Putnam quantitative lineage.	Recording and Chinese transcript remain unresolved.
Huabao Fund / Zhong Qi	CLS interview	A Chinese financial-media surface identifies Zhong Qi as an investment director and adds an academic-lineage lead for the mainland-China manager map.	Full article capture and translation remain pending; no model or deployment inference.
Citadel Securities / Costas Bekas	Google Cloud customer case	A vendor/customer surface names Bekas as Head of Research Platform and exposes a quantitative-research cloud-infrastructure partnership.	Vendor-reported architecture and outcomes; no model-level authority or performance evidence.
GTS Securities / Victor Zigdon	Google Cloud surveillance case	The case identifies Zigdon as Director of Trading Analytics and documents a GTS–Strike Technologies–Google Cloud surveillance pipeline joining proprietary order data with high-resolution market data.	Compliance/surveillance infrastructure, not alpha or portfolio-strategy evidence.
Arabesque AI / Nikolaos Kaplis and Matthias Baetens	Google Cloud case	A vendor/customer page adds named AI-organisation personnel and a dated investment-infrastructure surface.	Historical/vendor narrative; current architecture and results are not established.
UOB Asset Management / Paul Ho	Google Cloud case	A customer page adds an Asia-equities leader and a public case surface involving model testing and governance.	Vendor/customer evidence; no independent model-performance evidence.
MSCI / Hitendra Varsani and Andy DeMond	AI and risk transcript	The first-party transcript names investment-research and analytics-governance leaders and adds a title-blind AI/risk platform surface.	Institutional risk context, not hedge-fund deployment evidence.
Voleon / Blair Bilodeau	Personal research profile	A public researcher page adds a named Voleon research-staff signal and Toronto/Vector statistical-learning lineage.	Personal and academic evidence; no inference about Voleon’s current systems or performance.

Recovered direct audio and local ASR — August 18, 2026

The recovery pass resolved five direct media enclosures that the initial publisher-page review had not captured. The local transcripts are timestamped navigation aids; the public claims below are paraphrased and retain the source-specific boundaries in the capture note.

Firm / person	Recovered source	Publicly observable signal	Boundary
BlackRock Systematic / Raffaele Savi	Exchanges episode · recovered Megaphone audio	Savi discusses AI as a scaling layer for systematic investing, including more data and compute, language/text/image inputs, more interactive model interfaces, and possible extension into longer-horizon, less-liquid, private-market, and systematic-credit research.	Interview perspective plus direct audio and local ASR; no model registry, training corpus, live permission, or AI-attributed performance is disclosed.
Fundamental Edge / Portrait Analytics	Other People’s Money episode · recovered Megaphone audio	Caughran and Plon describe process decomposition, thesis monitoring, data-connected research workflows, deterministic APIs for portfolio/risk calculations, parallel manual and AI validation, and a boundary against outsourcing complete conviction-building.	Vendor/practitioner discussion and local ASR; no named customer deployment, permission map, evaluation result, or investment outcome.
Unnamed credit hedge fund / Sachin Kullkarni	Hedgineer Episode 4 · recovered Anchor audio	Kullkarni describes a translational data-science role and discusses Jupyter, Snowflake, Materialize, Airflow, structured/unstructured data, language-model routing, and deterministic risk APIs. The architecture example is a permissioned risk calculation explained through a language-model interface.	The current credit-fund employer is intentionally unresolved. The recording is practitioner evidence, not proof that the unnamed fund deployed the example or granted an agent trading authority.
Wellington Management / Mark Sullivan and Roberto Isch	Capital Allocators Episode 427 · recovered Libsyn audio · capture note	The recording describes Wellington’s multi-generation hedge-fund platform, manager underwriting, quantitative and qualitative research, portfolio construction, live risk monitoring, and a multi-asset risk model.	Public practitioner testimony and local ASR. It does not establish AI use, model ownership, permissions, or performance; the described process is not treated as an AI disclosure.
金声玉亮 / Lei Zong	Chinese-language Apple episode · recovered Xiaoyuzhou audio	The Chinese discussion distinguishes LLMs from time-series and other predictive models, and covers real-time data, retrieval quality, context contamination, multi-agent checking, and trust boundaries around automated or semi-automated investment assistance.	Chinese ASR still needs translation and speaker/timestamp spot checks. No current firm-system inventory or independently validated performance is established.

Title-blind Bloomberg quant-researcher media — Steve Hou, August 18, 2026

Searching for the named researcher and adjacent market topics—not for “hedge fund,” “quant,” or “AI” in the episode title—recovered three additional public episodes featuring Steve Hou. The publisher descriptions identify Hou as a Bloomberg quantitative or senior quantitative researcher covering multi-asset strategy research and holding a PhD in financial economics. The episodes cover AI’s effects on labor and markets, a public stock-index project and rebalancing, the US–China AI discussion, AI capex, compute demand, productivity, policy, and physical bottlenecks. The publisher pages and the local capture note now include recovered enclosures and timestamped ASR for all three episodes.

Public surface	Observable signal	Boundary
Full Signal — Quant explains how AI radically shifts the economy for investors · Podbean timestamps	AI, labor-market and market-structure discussion; the publisher description also names a Bloomberg index project and rebalancing.	Public media and researcher role description; not evidence of Bloomberg internal systems, permissions, or performance.
Full Signal — Wall Street MISSED this trade · Spotify	US–China AI discussion, geopolitical positioning, regional markets, and an AI-specific chapter at 37:06.	Public researcher commentary; no inference about a hedge fund’s AI deployment or investment authority.
Forward Guidance — The AI Bubble Is Widely Misunderstood	AI’s macro impact, historical bubbles, capex, agentic-AI compute demand, productivity, Federal Reserve policy, and bottlenecks.	Publisher timestamps and recovered public audio; the local ASR is an automated navigation aid requiring review before quotation.

This lane is adjacent market-research media, not a hedge-fund deployment record. It demonstrates why researcher-name, publisher, and title-blind searches must be run alongside keyword and RSS discovery. The evidence does not disclose a complete model inventory, data-permission map, internal Bloomberg deployment, investment authority, or independently measured result.

The same lane also recovers a personnel and academic bridge: MoneyShow’s public biography describes Hou’s prior systematic-equity research at AQR and a 2018 University of Michigan PhD; Michigan’s placement page lists the AQR placement, and Hou’s public CV names Stefan Nagel, John Leahy, Linda Tesar, and Paolo Pasquariello on the dissertation committee. This is a public career/academic lineage, not evidence that AQR or Bloomberg share a system, model, dataset, or investment result.

Regional first-party channel expansion — Kinea, August 18, 2026

The Kinea first-party blog links to a YouTube channel and Spotify show whose titles use investment themes, film references, and market commentary rather than “hedge fund,” “quant,” or “AI.” A title-blind crawl recovered three Portuguese videos and their auto-caption tracks. The capture note records the channel route and caption boundaries.

Firm / source	Publicly observable signal	Boundary
Kinea Investimentos	AI agents and the market states that Kinea has implemented its own agents and that research analysts are using natural-language coding to create code and analyze outputs.	Kinea-produced video and automated Portuguese captions; no model names, data permissions, evaluation results, or investment attribution.
Kinea Investimentos	US–China AI leadership discusses architecture, data, compute, model economics, and hardware constraints.	Investment and technology commentary; no Kinea model inventory or deployment permission is established.
Kinea Investimentos	AI infrastructure and value discusses semiconductor, energy, manufacturing, and physical-infrastructure bottlenecks.	Public market commentary; no portfolio position or model use is inferred.

Title-blind allocator-media expansion — AI for Allocators, August 2026

The PodcastIndex crawl found a second, previously untracked route: the AI for Allocators feed, which contains ten episodes published from June through September 2025. The titles do not identify hedge funds, quant research, or specific firms. The series’ own recovered audio says it is AI-generated from Shaun Wei Tjia Ng’s original blog posts. That production detail matters: the episodes are useful as a first-party allocator-framework and vocabulary surface, but they are not human guest interviews and do not establish that any investment office adopted the proposed workflows.

Surface	Publicly observable signal	Boundary
Moving too fast on AI—or too slow?	Frames allocator AI adoption as a change-management and capability-building question, with explicit attention to security approvals, focused pilots, culture, and investment-committee impact.	AI-generated series content; no named office, deployment, budget, or measured result.
The Leader’s Playbook and Five High-Impact AI Habits	Proposes shared prompt libraries, leaders modeling imperfect outputs, and protecting time created by automation for deeper analysis.	Proposed operating practices, not evidence of adoption by a specific firm.
Three Prompts Every Allocator Should Run	Describes style-drift checks against historical manager letters and memos, return-driver and hidden-risk review, and triage of which pitch decks merit meeting time.	Proposed diligence prompts; no validation, data-permission map, or investment attribution.
Five Overlooked Ways AI Is Impacting Investment Offices and Allocator’s Underutilized AI Toolkit	Names manager diligence, research, onboarding, deep-research agents, citation checks, false-negative risk, and enterprise data-security boundaries as distinct workflow questions.	Series framework; no named customer, model inventory, or outcome evidence.
Four No-Regret Moves, Busy Week in AI, and Three High-Impact Actions	Calls for tagged historical documents, open APIs, bring-your-own-model support, granular agent permissions, audit logs, and model portability; it also discusses GPT-5, Claude Opus, and open-source models as dated market commentary.	Commentary does not establish that the named tools are deployed in an allocator or that vendor claims were independently tested.

The capture note contains the complete ten-episode inventory, direct enclosures, timestamps, and raw-capture boundary. The episode sidecars are timestamped machine ASR with no speaker diarization and are retained for search rather than quotation.

Written-source branch — AI for Allocators on LinkedIn

The series is larger than its podcast feed. Its LinkedIn newsletter and author profile expose dated written sources that add implementation vocabulary and vendor-discovery leads. These are first-party recommendations and frameworks, not disclosures by the allocator offices discussed in them.

Written source	What it adds	Boundary
Small Family Offices & AI — What Comes Next	A problem-statement, budget, and market-engagement sequence; it names Clade, FinPilot, Vantager, and Hebbia as examples of vertical tools for manager selection, workflow, and private-asset research.	Author’s dated vendor list; not a product audit, endorsement, customer deployment, or performance claim.
Choosing an AI Vendor	Short contracts, clean exits, model-agnostic architecture, traceability, and a proposed 500–1,000-document proof-of-concept floor.	Author guidance; the document threshold is not an independently validated benchmark.
Mapping the AI Touchpoints Across Your Investment Office	Search terms beyond models: governance, talent, operational due diligence, cyber/deepfake risk, strategic asset allocation, and team literacy.	A taxonomy for discovery, not evidence of a named firm’s system.
Designing the Environment that Propels the AI Flywheel and What Matters Most?	Living policies, approved and prohibited use cases, shared learning, leadership engagement, and leverage-point questions for manager selection, portfolio construction, relationships, and stakeholder communication.	Operating guidance; no model, permission, data, or return disclosure.
Twelve Months of Conversations	States that the series is grounded in private conversations with allocators and frames adoption as leadership, culture, and workflow redesign.	The underlying conversations are not public and cannot be independently attributed to firms.

The author’s LinkedIn posts also link to a YouTube conversation. The route is now resolved: YouTube identifies it as AuumAI’s “Shaun Ng on the LP AI Misdiagnosis and Building a Thriving Investment Office in an AI World,” published June 16, 2026, with English automatic captions and a 1,378-second runtime. It is a mirror of the Shaun Ng interview already captured from the publisher’s direct audio enclosure, so it adds a searchable timestamp surface rather than a new independent finding. The captions were not manually reconciled and are navigation evidence, not a verbatim quotation source. The expanded capture note records the provenance and deduplication boundary.

What changed in the evidence map

1. RSS was a transport layer, not a discovery layer

The earlier process over-weighted registered RSS feeds. The new pass found material on official firm YouTube channels, Bloomberg publisher pages, Acast show pages, Castos transcripts, HFR’s media index, and speaker-name searches. PodcastIndex is now useful for feed and episode graph discovery, but its transcript grep is scoped to published transcript files in selected feeds rather than a universal episode-text index.

2. Guest-name search recovered the missing personnel

Searching “AI” and “hedge fund” alone buried the relevant items. Searching Iain Dunning, Mike Schuster, Matthew Granade, Nishant Gurnani, Jon McAuliffe, Jean-Philippe Bouchaud, and Richard Craib recovered sources that generic topic filters did not surface.

3. The media divides into four evidence lanes

Lane	Sources in this pass	Observable content
Predictive-market ML	HRT, Jane Street, Voleon, InfoQ	Market data, model research, compute, latency, database, and validation constraints.
Research workflow and agents	Balyasny, Versor, Numerai, Point72	Retrieval, coding, hypothesis generation, evaluation, and human review.
Scientific/systematic lineage	CFM, Two Sigma, Voleon	Research culture, statistical modeling, feature work, and academic/technical lineage.
Industry/QIS context	HFR, InfoQ, broader macro interviews	Method vocabulary and market-wide framing without firm-specific deployment proof.

These lanes should not be collapsed into a single “AI maturity” score. A predictive-market system and a research assistant answer different questions and require different evidence.

Title-blind discovery expansion

The intake system now treats episode titles as one weak signal rather than the search boundary. RSS records are matched against title, description, subtitle, author, tags, and publisher text. YouTube sources can also run targeted guest, role, firm, and vocabulary searches alongside channel listings. The registry now carries queries such as “quantitative researcher,” “head of research,” “head of automation,” “chief technology officer,” “alternative data,” “market microstructure,” “earnings calls,” “GPU,” “data platform,” “research agent,” and “portfolio construction.”

This matters because the useful disclosure may be titled “Finding Signal in the Noise,” “The AI Supply Chain,” “Inside Data Science,” “The Quant Edge,” or a person’s name. The absence of “hedge fund,” “AI,” or “machine learning” from a title is no longer treated as negative evidence.

The title-blind queue also recovered the Excess Returns interview with Michael Robbins, published September 22, 2023. Robbins discusses model specification before fitting, alternative-data scale, machine learning for asset allocation, risk and backtest design, possible manager-allocation models, and causal graphs that can flag broken assumptions (04:27–06:24; 07:13–08:36; 30:57–31:12; 47:59–50:36). He also uses facial expressions during earnings calls as an example of a possible voice/vision signal (12:27–12:39). This is a dated academic/practitioner methodology source, not evidence of a named hedge fund’s deployment. Robbins’s first-party media index exposes additional QuantVision, MathWorks, ReSolve, Blockworks, Risk.net, and allocator-media routes for systematic follow-up. See the capture note.

The same queue recovered a March 29, 2026 TraderLion interview with Jim Roppel, whose title does not mention AI. The episode’s publisher transcript places an AI discussion at 01:41:21: Roppel describes AI as a market theme, a limited personal use of an AI tool for historical price-pattern questions (01:44:34–01:45:08), and a view that human psychology remains relevant to price formation (01:49:50–01:51:10). This is manager commentary and a personal workflow anecdote, not evidence of a firm model, production deployment, or AI-generated investment signal. The capture note records the automatic-caption boundary and distribution variants.

The next search surfaces are now explicit intake lanes:

Surface	What it can reveal	Treatment
Firm-controlled podcasts and engineering channels	Research workflow, compute, hiring, data systems, production constraints	Primary practitioner evidence when the speaker and date are clear
Quant and allocator shows	Factor design, alternative data, portfolio construction, risk, and implementation vocabulary	Source discovery until linked papers, code, or official material is archived
Vendor customer sessions	Named data platforms, model providers, cloud services, and workflow architecture	Separate customer statements from vendor framing
Conference recordings	Keynotes, panels, and technical sessions that use generic titles	Search speaker names and session descriptions, not only event names
Job and personnel surfaces	Titles that imply ownership without using “AI”	Date-scoped personnel evidence; never infer deployment from a title alone
Papers, GitHub, competition notebooks, and technical blogs	Model families, datasets, evaluation designs, and implementation details	More reliable corroboration when the artifact is attributable and reproducible

The registry and miner changes are recorded in the source ledger. This is an expansion of recall, not a claim that every discovered item is relevant or verified.

Two recurring index surfaces are also queued: AQR’s official press index, which points to researcher interviews and podcasts, and YouTube’s Researcher Program, which may provide broader public-video metadata access for eligible researchers. Neither is treated as transcript evidence until the linked source is independently reviewed.

P0 person-level sweep — August 19, 2026

The next person-level pass recovered official and publisher surfaces that generic “hedge fund AI podcast” searches did not reliably return:

Person / firm surface	What the public source adds	Capture boundary
Iain Dunning — HRT AI Labs	NVIDIA’s GTC page identifies Dunning as Head of HRT AI Labs and describes tick-data properties, transfer limits from deep learning, and reinforcement-learning questions in quantitative trading. HRT’s LinkedIn post supplies the dated event route.	Official session abstract and firm announcement; the linked YouTube route resolves to an unrelated MMTLP short, so it is not evidence of the talk recording.
Ben Wellington — Two Sigma	Two Sigma’s own summary names Wellington as Deputy Head of Feature Forecasting and discusses feature construction, earnings-call metadata as a hypothetical example, data saturation, interpretability, and timestamp integrity.	First-party summary plus the already captured podcast route; hypothetical features are not treated as live model disclosure.
Igor Tulchinsky — WorldQuant	The Voices of Impact episode page dates a 20-minute interview to May 28, 2026. Tulchinsky’s LinkedIn post says it covers education, WorldQuant University, and how AI changes participation in quant finance.	The World Jewish Congress episode page exposes a separate public S3 MP3; local aligned ASR now locates his “data intelligence unit” description (10:45–11:14) and one-person-plus-AI IQC claim (15:33–15:52). These remain founder-reported, ASR-qualified signals without model inventory, deployment, permissions, or performance proof.
Bryan Kelly — AQR/Yale	Yale’s current profile identifies Kelly as head of machine learning at AQR and lists financial machine learning, asset pricing, volatility, tail risk, correlation, and financial networks.	Current academic/role evidence; an older AQR video URL now returns 404 and is retained as a locator failure.
Jean-Philippe Bouchaud — CFM	The April 2026 Risk.net article links model-guided GenAI to scarce, non-stationary financial data, overfitting, path dependence, and a CFM ML Lab collaboration on “moment guided diffusion.”	Trade-publication and CFM-associated research narrative; no production or performance evidence.
Petter Kolm and Nicholas Westray	The public paper page and Risk.net follow-up expose research on four neural architectures, limit-order-book data, timestamping, input construction, and lookback design.	Research and author lineage; no named-firm implementation or full-paper capture was promoted.
Vasant Dhar — Brave New World	The official RSS feed exposes the latest ten posts, while the public WordPress REST collection reports 110 posts across two pages. A title-blind filter recovered canonical episodes on investing, forecasting, AI alignment, governance, human-centered AI, machine-mediated groups, and AI-native product strategy. The Thinking With Machines recording now adds automatic-caption navigation for Dhar’s historical trading-algorithm and machine-learning discussion (12:00–16:03) and his distinction between broad market prediction and repeated human-behavior patterns (18:11–20:58). A separate 2026 book interview adds adjacent finance, judgment, and human–machine decision context. Dhar’s NYU profile connects the show to data science, AI trust, and historical SCT Capital Management.	Archive, episode metadata, and public caption routes are verified. Captions are automatic and not treated as verbatim transcripts; the recordings establish no current SCT system, model inventory, data rights, deployment, or investment result.
Tucker Balch — JPMorgan AI Research / Georgia Tech	The seminar page covers ML and multi-agent simulation for market-rigging questions, synthetic data in finance, and computer vision for forecasting. Georgia Tech’s ML-for-Trading course adds the educational lineage.	University event and course evidence; no JPMorgan model inventory or production permission.

The full source note and ledger entries are in the P0 person-level capture note, the Brave New World archive note, and the cross-dimensional coverage ledger.

Zero-coverage and title-blind firm sweep — August 19, 2026

A second pass targeted firms that had zero or one verified media record. It found dated first-party or institutional surfaces for Renaissance, Millennium, PIMCO, Dimensional, Aspect, QRT, Squarepoint, Voleon, and Arrowstreet:

Firm / route	What the source establishes	Boundary
Renaissance Technologies — Peter Brown	Goldman Sachs identifies Brown as CEO and describes a 2023 interview about the firm’s history, crises, and the long-running role of computer models and algorithms.	Official podcast context; no current model inventory or technical transcript was archived.
Renaissance title-blind route: Acquired and searchable transcript	A March 2024, three-hour episode whose title contains neither “AI” nor “quant.” The secondary discussion covers early Markov/HMM framing, data cleaning, IBM speech-recognition talent, machine-discovered relationships, human review, and model overrides at dated transcript timestamps.	Promoted as secondary historical context and a discovery route, not as a Renaissance disclosure. The transcript and hosts’ interpretation require audio spot-checking and do not establish current architecture, personnel, permissions, or performance.
Michael R. Douglas — CBMM/MIT video and Jade Vinson — Stony Brook video	University-hosted videos provide dated Renaissance affiliations and searchable/downloadable academic talks: machine learning and mathematical research for Douglas; geometry for Vinson.	Historical personnel and academic-topic evidence only; neither page discloses a Renaissance investment system or current role.
Millennium — Gideon Mann and technology page	The firm identifies Mann as Global Head of Artificial Intelligence, Technology, describes an AI advisory group, and publishes a technology/AI platform narrative.	First-party role and strategy language; no system ownership, model list, or performance evidence.
Millennium Dublin ML event	An institutional event report names Paolo Aloe and Pat Lenihan and describes natural-language SQL, multi-step research agents, and cross-model LLM standardization.	Event-report evidence; no recording or firm-wide deployment map.
PIMCO — Andrew Balls	The timestamped interview describes AI for credit-document and call analysis, unstructured data organization, sentiment, and forecasting tools.	Named executive account; no audited model or direct-alpha attribution.
Dimensional — The Informed Investor	The firm’s Episode 13 identifies Mark Gochnour, Jake DeKinder, and Wes Crill in an AI-and-investing discussion.	Firm media and portfolio framing; no internal AI lab or production model disclosure.
Dimensional — Weston Wellington · Apple listing	The May 19, 2025 Talking Real Money episode has no AI term in its title, but its publisher show notes place “AI vs. aggregated intelligence” at 21:01 and identify Wellington of Dimensional Funds.	Named-person and title-blind media evidence; no Dimensional model, AI system, or production-workflow disclosure.

The title-blind Dimensional pass adds several adjacent routes that were not visible in the AI-titled search. Rob Harvey’s April 27, 2025 Behind the Ticker episode identifies him as a Dimensional vice president and links the same episode to YouTube, Buzzsprout, and Spotify. The public show notes discuss evidence-based ETF construction, daily portfolio management, and the research bar for portfolio changes. They do not disclose a Dimensional AI system, model, training corpus, or agent authority. Orion’s episode with Savina Rizova identifies her as Co-CIO and Global Head of Research, gives her public academic lineage, and places big data, AI, and machine learning in the outline at approximately 28:20–31:37. The Gerard O’Reilly episode similarly identifies him as Co-CEO, Co-CIO, and former Head of Research, with AI and innovation placed at approximately 32:10. These are publisher and executive-context signals; the pages do not establish implementation details, model ownership, data rights, or production deployment. See the capture note. | Dimensional — Wei Dai · Spotify show · Apple show · YouTube channel | The April 1, 2026 Singapore publisher episode identifies Wei Dai as Dimensional’s Global Head of Research and discusses systematic investing, factor research, diversification, and market-prediction limits. | Regional, title-blind publisher metadata; the linked audio enclosure was not recoverable in the check, and the page makes no AI-system or deployment claim. | | Aspect — Martin Lueck | Aspect’s own summary links Lueck’s podcast appearance to scientific investing, hypothesis testing, model shutdown, machine learning, ChatGPT/LLMs, and risk management. | Topic summary; full audio and transcript remain uncaptured. | | QRT — NUS technical event | The university event describes QRT participation across research, technology, recruiting, data science, and machine learning. | Names no speakers and exposes no recording or system detail. | | Squarepoint careers | Squarepoint’s current page explicitly connects machine learning with quantitative models, automated trading, research platforms, and trading infrastructure. | Current firm wording; no named personnel, model, dataset, or result. | | Squarepoint ICML sponsorship and early-careers page | The dated public post identifies ICML sponsorship and named recruiters; the careers page connects ML degree backgrounds to investment internships and data-intensive technology work on research and trading systems. | Recruiting and conference-presence evidence; no named AI lab, paper, model inventory, or production/performance claim. | | Squarepoint personnel and research-lineage routes | Public CVs and profiles connect Squarepoint personnel or interns with optimizer dynamics, conic portfolio optimization, ML-trading-model interpretability, reinforcement learning, causal imputation, and generative-model research. | Individual and historical artifacts; they do not establish a current firmwide research program or that the academic methods are deployed at Squarepoint. | | Voleon official management and ML pages | Voleon’s own pages name technical leaders and describe machine-learning-based financial prediction, portfolio optimization, market-impact estimation, and regional research responsibility; the public biographies also expose Stanford, Berkeley, Princeton, and CMU lineages. | First-party biographies and strategy wording; no complete model inventory, dataset, permission map, or performance attribution. | | Jon McAuliffe — Masters in Business | The title-blind publisher page identifies the Voleon co-founder/CIO and describes a four-stage process of data assembly, prediction engine, portfolio construction, and execution, with Apple, Spotify, YouTube, Bloomberg, and transcript routes. | Named executive account and show notes; independent verification and audio/transcript spot-checking remain open. | | Voleon academic routes | Stanford’s 2018 event page identifies Voleon senior research staff member Mike Ryerson and a talk on the benefits and implementation of machine learning for quantitative investing; Berkeley’s Voleon seminar index exposes a wider network across deep learning, causal inference, RL, value of data, and human-AI information. | University event and sponsor-network evidence; no claim that Voleon deploys each seminar topic. | | Voleon researcher lineage | Blair Bilodeau’s profile connects a Voleon research role to statistical/ML methods for systematic investing, a Toronto PhD advised by Daniel Roy, Vector Institute affiliation, Google Brain experience, and NeurIPS/ICML research. | Public personal profile; it does not disclose Voleon systems or permissions. | | Arrowstreet technology | Arrowstreet describes scalable platforms, advanced data science, high-performance computing, and technology working across investment and business teams. | First-party technology description; no named AI leader, vendor, or GenAI system. | | Arrowstreet homepage | The current firm homepage reports $344B+ AUM, 440+ clients, and 500+ employees as of June 30, 2026, while describing quantitative signals, large datasets, next-generation platforms, advanced data science, and high-performance computing. | Dated first-party scale and operating-surface evidence; not an AI-capability, model-ownership, permission, or performance measure. | | Arrowstreet Quantitative Researcher posting | The current role description names signal codification, backtesting, return/risk/trading-cost forecasts, structured and unstructured data, simulation, third-party feeds, data architecture, production code, and LLMs/coding agents as a plus for research and programming workflows. | Role-level hiring evidence; no filled-role, model, vendor, corpus, autonomous-agent, or performance claim. | | April Rathe — PyData | The conference profile identifies Rathe as leader of Arrowstreet’s Research data-science platform and links a “Dask: From POC to Production” presentation. | Dated personnel and platform-conference evidence; presentation capture remains open and does not disclose investment models. |

The full capture note is the zero-coverage and title-blind sweep.

Newly discovered but not yet promoted

The search also found useful leads that remain below the promotion threshold:

• The Acquired Renaissance episode is now promoted as secondary historical context; it remains unsuitable as a current first-party disclosure until audio spot-checking is complete.
• The Andrew Beer / Dynamic Beta Investments episode and its Podscan transcript route list an AI-in-portfolio-research segment at 56:00 alongside systematic investing, managed futures, and hedge-fund replication. The full authorized MP3 is now transcribed locally with timestamped English ASR. Beer says AI improved programming efficiency without changing the core investment idea (57:21–57:52) and describes using Claude for broad allocator research on diversification and managed futures (57:52–59:39). The account is speaker-reported, not a model-deployment or AI-alpha claim; no diarization, data-rights, permission, or performance evidence is inferred. It is tracked in the independent-publisher source note.
• Two additional episode-level routes were found by searching the event narrative rather than a firm’s AI vocabulary: Market Maker on Spotify and The Wall Street Skinny on Podscan. Both discuss the 2026 AI-market unwind and related financing/liquidity questions. They are registered as secondary commentary and deliberately not promoted as evidence of any named firm’s internal AI strategy, model, data rights, permissions, authority, AUM, leverage, transaction, or performance. See the source note.
• The same story-specific sweep recovered Bloomberg’s Big Take, Pivot to AI, and The Future of Work with Jacob Morgan. These are registered as additional secondary routes and remain deliberately unpromoted; their editorial accounts are not substitutes for filings, firm statements, or directly attributable system evidence. See the expanded source note.
• QRT and Squarepoint searches surfaced secondary references and talent-market signals but no new canonical practitioner transcript in this pass.
• The Squarepoint pass now promotes primary firm and conference routes plus individually attributable public CV/profile routes. A secondary industry case study that describes a Squarepoint data-science organization, alternative-data signals, reinforcement-learning review, and an internal GenAI platform remains a lead-only record until a Squarepoint-controlled source corroborates it.
• The Voleon pass adds a primary management/ML record, a title-blind executive podcast route, a Stanford machine-learning-for-quant-investing talk, Berkeley’s recurring Voleon seminar archive, and named researcher lineages. The podcast and seminar routes are kept distinct from firm-controlled system evidence.
• The Arrowstreet pass found a role-level signal: a current Quantitative Researcher posting explicitly lists LLMs and coding agents as a plus while describing production research code, signal backtests, simulation, unstructured data, third-party feeds, and data architecture. It also identified April Rathe as a public Research data-science-platform leader through PyData. These are workflow and talent signals, not evidence of a named production GenAI system.
• The GMO pass adds two previously unpromoted title-blind practitioner routes, a GMO-hosted emerging-markets transcript, and public AI-investment conference materials. It now separates GMO’s AI-as-an-investment thesis from internal-AI evidence and preserves Hancock’s historical ML/academic lineage and Chiang’s systematic-equity process discussion as distinct records.
• The Systematica pass adds Goldman’s 2019 ML-trend interview, Portuguese-language Outliers/InfoMoney routes, a 2026 SBAI operational-risk episode, current Data R&D and Shanghai modeling-language hiring signals, and a historical production-systems profile. These widen coverage across research, regional media, governance, and infrastructure without merging them into one AI-deployment claim.
• The earlier negative-search line for Voleon is corrected: a full 2023 publisher transcript exists, and a linked YouTube automatic-caption track now supplies timestamps. The source ledger distinguishes “not found in the earlier pass” from “no public evidence exists.”
• Man Group’s July 2026 token-spend episode is now locally archived and promoted in a dated note; its 86× consumption figure remains a firm-side claim without an audited baseline, absolute spend, or outcome attribution.
• Timefolio’s Singapore CIO on J.P. Morgan’s “AI supply chain” episode now has a first-party transcript, recovered Brightcove audio, and timestamped local ASR. The episode is useful for the distinction between AI as an investable supply-chain theme and AI as an internal investment-process system: Lee describes a discretionary/fundamental workflow using supply-chain, import/export, and scraped data, while the host explicitly says this episode is about the theme rather than the usual “AI in the investment process” format. No GenAI deployment claim is promoted. Evolution Exchange Singapore’s “AI in Hedge Funds” episode has cleared the canonical-page check and Acast enclosure recovery; its transcript is an internal ASR layer pending manual name/timestamp spot checks.

Generating Alpha title-blind executive expansion — August 20, 2026

The Generating Alpha publisher surface was searched by firm leader and founder names before applying AI-related keywords. That recovered four episodes absent from the canonical media ledger. The full cross-platform capture is retained in the repository’s source ledger; the published evidence is linked directly in the table below.

Firm / person	Recovered source	Publicly observable signal	Boundary
PDT / Pete Muller	Generating Alpha Episode 47 · YouTube captions	Around 14:40–15:12, the discussion emphasizes statistical-model use and overfitting risk; around 23:43–24:02, it frames quantitative decisions as trust-or-modify model decisions; around 24:22–24:33, it places large-model discussion beside established quantitative practice.	Third-party interview and automatic captions; no PDT LLM, agent, data, permission, deployment, or performance disclosure.
Balyasny / Dmitry Balyasny	Generating Alpha Episode 48 · YouTube	Around 30:38, the recovered material discusses collaboration across credit, macro, commodities, and quantitative research. A later AI reference is retained as a lead pending cleaner transcript verification.	Executive interview; no new model inventory or deployment claim. Specific Applied AI architecture remains attributed to Balyasny/OpenAI and first-party sources already cited above.
Schonfeld / Ryan Tolkin	Generating Alpha show index · YouTube captions	Around 36:37, the conversation turns to AI and Schonfeld’s AI efforts. Around 42:42–42:52, it connects data materiality and AI to a research environment where data and research are increasingly available.	CEO/CIO interview with automatic captions; no named model, evaluation result, permission map, deployment stage, or return attribution. SchonAI and partnerships remain grounded in the first-party FE AI Lab disclosure.
Susquehanna / Jeff Yass	Generating Alpha Episode 39 · YouTube captions	The conversation covers prediction markets, manipulation and systemic risk, and the limits of deciding what should be quantified. It is decision-science context for quant research.	Founder interview; no Susquehanna AI system or investment-model disclosure.

These records add a reusable search rule: enumerate publisher feeds by firm leaders, CIOs, CTOs, heads of research, and founders, then resolve Apple, Spotify, YouTube, and transcript surfaces. A missing AI keyword in an episode title is not a negative finding.

Susquehanna: deep-learning leadership and a public quant role taxonomy

The title-blind queue also recovered a Susquehanna firm-published quant hiring video, published July 9, 2025. The firm’s LinkedIn post identifies the speaker as Lyubo Panchev, a quantitative researcher and hiring manager. The video describes quants as building mathematical frameworks over market data and says the work can span mathematics, statistics, computer science, machine learning, finance, data science, and game theory. Around 05:43–06:59, Panchev distinguishes quant roles centered on machine-learning developments or deep-learning infrastructure from roles closer to strategy and trading. Around 08:34–10:39, he describes an iterative process of forming an expected relationship, testing it against data, and investigating missing variables. Around 13:58–15:24, he separates research, developer infrastructure, and trader oversight while noting that the allocation of responsibility varies by asset class and system.

Susquehanna’s ICML 2026 page adds a current public research and recruiting surface. It lists Hailin and Duccio as machine-learning researchers, Illan and Davor as quantitative researchers, Shirin as a machine-learning researcher, and Ali Nazari as Head of Deep Learning Research. The page connects machine learning and advanced quantitative research to large datasets, systematic trading, and collaboration among researchers, engineers, and traders. A firm LinkedIn post describes a Diamond Sponsorship, machine-learning and quant lightning talks, and an ROIbot trading game. The firm’s public feed also announces an Agentic AI Summit 2026 panel with Nazari, Jeff Wecker of Two Sigma, Jen Allum of D. E. Shaw, and Li Deng of Vatic Investments.

These sources support named leadership, a public quant/ML role taxonomy, and a dated conference and recruiting presence. They do not disclose a model inventory, training corpus, benchmark result, agent permissions, production endpoint, portfolio authority, or AI-attributed performance. The video captions contain proper-name errors, so the timestamp links are navigation aids and the analysis is paraphrased rather than treated as a clean transcript. The full capture note preserves those boundaries.

Canadian quant-media follow-up — August 20, 2026

The Canadian-tilt audit recovered two media routes that were absent from the ledger. The Mackenzie Investments episode features Arup Datta, Head of the Global Quantitative Equity Team, and was recorded March 2, 2026. Its publisher description emphasizes transparency, governance, human oversight, global stock selection, and risk controls. A separate Institutional Connect publisher post and transcript adds practitioner language about alternative data, text parsing, inference, machine learning, and NLP. The Castbox page does not expose a full transcript, so the claims are kept separate by source type.

The Index Fund Advisors video identifies Wes Crill, a Dimensional Fund Advisors Vice President, and dates the conversation to August 29, 2023. It is third-party video metadata about AI in investing, not evidence of a Dimensional model, internal deployment, or portfolio permission. The full capture note is retained in the repository ledger rather than linked as a public page.

Fidelity Investments Canada’s QRI adds a Canada route found through a title-blind regional search. In a first-party May 2026 transcript, Karishma Kaul describes QRI’s systematic fixed-income research workflow, its roughly 200 quants and 250-plus technologists, centralized data controls, alternative-data inputs, model and backtest review, transaction-cost modelling, and research-approval process. She reports that AI tools made the team two to three times more productive in research validation, new-data ingestion, and model review. That figure is participant-reported and has no disclosed time-study denominator; the public record does not establish model vendors, training data, licenses, portfolio permissions, or AI-attributed returns. The Spotify route cross-checks the episode identity.

J.P. Morgan Asset Management’s Hong Kong-based On Investors’ Minds APAC archive adds three title-blind regional records. Episode 163 identifies Fiona Harris and describes a workflow in which analysts write predetermined company game plans while AI analysis reviews emails and meetings for thesis drift and may compare a thesis with peers and competitors (09:29–12:04; capture note). Episode 160 discusses CFO scrutiny of token or volume costs and reserving expensive models for higher-value tasks (02:15–03:02; capture note). Episode 161 supplies a fixed-income portfolio-manager route but no internal AI-workflow disclosure. The publisher and local audio captures identify the roles and dated statements; they do not disclose model providers, training data, permissions, retention policy, evaluation design, portfolio authority, or AI-attributed returns.

Bernstein’s Rupal Agarwal episode adds a Singapore/Asia quant-research and event route that was missed by the earlier hedge-fund-only searches. Bernstein’s personnel page identifies Agarwal as Asia Quantitative Strategist. The April 2026 episode describes agentic workflow expansion, unstructured-data research, governance, role-specific training, and human escalation; it also names examples involving management-narrative and sentiment analysis, voice and text patterns, scenario generation, and a vendor innovator series including Anthropic, Fiscal AI, Boosted AI, Delupa, AlphaSense, Rogo, Hebbia, LinqAlpha, Finster, and QuantConnect (01:58–06:41; 11:58–19:23; 24:15–27:32; capture note). Bernstein’s first-party symposium post names Agarwal’s Gen AI in Asset Management: Intelligent Capital Blackbook and participants from BlackRock, Tenucia Partners, LinqAlpha, and Model ML. These sources establish public research and event disclosures, not procurement, vendor partnerships, internal model ownership, production permissions, or performance.

UK systematic-media follow-up — August 20, 2026

The UK thin-lane pass recovered two additional routes. A University of Cambridge seminar page identifies a 2021 GSA Capital technical talk by Joris Peeters. Its abstract discusses new data streams, Reddit, AWS simulations, latency, and technology teams; no recording or transcript was recovered, so it remains event metadata rather than current AI-system evidence.

The Top Traders Unplugged episode with Simon Judes provides a timestamped publisher transcript. Around 05:05–08:26, Judes discusses empirical research, multiple-hypothesis controls, and avoiding private overfitting; around 35:00–38:16, he discusses machine-learning data demands and the limited number of independent observations at slower trading horizons. The episode is historical executive research-process evidence, not a current Winton model inventory or deployment disclosure.

Chinese-language quant-video and roadshow expansion — August 20, 2026

The regional-language pass recovered five routes that English-only searches did not surface. Lingjun CIO Ma Zhiyu appears in a Douyin/China Securities Journal video and a separate China Fund News forum video. The latter’s public MP4 has now been recovered and processed with local automatic Chinese ASR, adding timestamped navigation for Ma’s general framing of AI as part of ongoing investment-technology iteration (00:45–01:11). The short statement does not add a model, dataset, vendor, permission, or production claim; see the capture note. DeepWin’s Wei Mingsan roadshow provides visible chapters and a described workflow in which research, coding, evaluation, and fund-manager review roles are separated. The page is a firm-hosted/publisher roadshow, not an independent audit. A historical Mingshi Bilibili video adds third-party Chinese-language quant/AI context without a verified transcript. Finally, a WizardQuant/WILL recruiting post names AI research, LLM, RL, MLSys, and large-model application roles in Shanghai and Beijing. These records are retained with original-language and translation boundaries; none establishes model weights, filled roles, live permissions, or AI-attributed returns.

Japanese and Chinese title-blind asset-management media — August 25, 2026

A Japanese first-party interview from Sumitomo Mitsui DS Asset Management, published February 20, 2026, identifies investment-development group head Hirose Takehide / 廣瀬勇秀 and describes AIR, an internally developed Assistant For Investment Research. The company says internal use began in October 2025 and describes AIR as combining multiple cloud generative-AI models with purpose-built agents on a closed internal network. The interview names Azure, AWS, and Google Cloud as selectable model environments and describes agents for research collection, summarization, long-report drafting, fact checking, and news/price analysis. Hirose frames the tool as support for idea generation, hypothesis testing, and risk-scenario review. This is unusually specific first-party implementation evidence, but it does not disclose model versions, data permissions, evaluation sets, agent tool scopes, portfolio authority, or AI-attributed performance.

The Japanese Amova robotics interview was reviewed in the same pass. It discusses AI training, inference, agents, physical AI, and a robotics investment theme, but it is not evidence of an Amova/Nikko internal quant or GenAI platform. It is retained as a title-blind company-media false-positive control.

A Simplified-Chinese Lujiazui Financial Salon report, published July 23, 2026 after a July 19 roundtable, names Higgs Investment Management chairman/general manager Tan Xiaojun / 谈效俊, University of Science and Technology of China professor Yao Jiaquan / 姚加权, Daguan Data chairman Chen Yunwen / 陈运文, and a Huawei finance-sector expert. The discussion describes data quality, ontology, source verification, constrained rules, agent stages from research assistance to restricted execution, and the need for out-of-sample/live consistency, full audit trails, and explicit risk-budget authorization. It is regional financial-media reporting of a roundtable—not a Higgs model card or production audit—and does not establish that every described stage is live.

A separate 10jqka / China Securities Journal survey reports that a Mingshi representative described current AI-agent use as concentrated in AI-assisted programming, with core factor mining still using existing methods and the deployment as exploratory. That is a useful disqualification boundary: an AI coding assistant or experiment should not be relabeled as a live factor-discovery or portfolio-execution system. The full source note, including original-language and capture metadata, is here.

A title-blind Xueqiu-produced Hou Xue Chang Bo episode, also distributed through Podimo and a Zeno.FM publisher mirror, identifies Chen Peng / 陈鹏 as a partner and macro-quantitative investment manager at Yuanlan Fund / 远澜基金. The distribution pages place it in June 2026. The episode title does not mention quant or AI, but its chapters discuss AI-assisted strategy iteration (26:16), reading news and converting text into judgments (27:51), differing responses to similar geopolitical events (37:17), and a digital-virtual-person stock-selection experiment that remains in the laboratory (47:56). The newly recovered private Chinese ASR adds timestamped navigation: the speaker describes shortening a strategy-development cycle from about a week to about a day and broadening exploration (27:19–27:48); replacing narrow keyword rules with model-assisted global-news interpretation (28:15–29:08; 29:38–29:53); and a digital-discretionary-trader concept whose consistency, explainability, and traceability remain unresolved (47:56–51:02). The recording also contains an unverified speaker claim about a fully AI-driven model operating for more than six months; it supplies no return series, risk statistics, benchmark, or reproducible evidence, so it is not treated as a performance result (57:24–58:07). This is publisher/guest testimony, not a model card or independent evaluation; no model identity, training data, permissions, or portfolio authority is inferred. The original M4A and 2,149-segment timestamped Chinese ASR are privately retained; see the updated capture note.

Fresh regional and conference gap-pass — August 26, 2026

The FIAM Montréal Summit 2026 program is scheduled for August 27, 2026 and includes sessions on AI in asset management, governance and deployment, and “The Rise of Agentic Finance.” The published roster names Gerald Garvey of BlackRock Systematic, Anne-Sophie Van Royen of Northern Trust Asset Management, Katia Walsh of Apollo, Gilbert Haddad of Fidelity, Jérémie Canoni-Meynet of DCM Systematic, and McGill professor Russ Goyenko. The organizer’s biography for Haddad also describes prior Point72 and Balyasny roles, while the biography for Garvey lists academic appointments in Canada and Australia. These are dated program and biography records; the page is not a recording or transcript and does not establish that any named firm uses a shared system, that a described agent has investment authority, or that any result is attributable to AI.

The Optiver research page describes a proprietary-trading research lifecycle spanning problem definition, data selection, hypothesis testing, simulation, failure analysis, and conversion of models into systems subject to live constraints. It names an AI Lab and lists deep learning, reinforcement learning, LLM-based analysis, signal discovery, forecasting, execution quality, and risk decisions among its public research areas. A separate Optiver “AI on a live trading floor” event page dates a July 22, 2026 Amsterdam event and describes foundation-model work across research, coding, testing, and live-system improvement. Both pages are first-party positioning and event evidence; neither supplies model weights, vendors, training data, permissions, deployment telemetry, or performance attribution.

The India-focused CafeMutual interview with Bhautik Ambani, published May 18, 2026, identifies Ambani as CEO of AlphaGrep Mutual Fund and describes intended AI support for data analysis, signal generation, risk monitoring, and anomaly detection within regulatory boundaries. The mutual-fund vehicle is kept separate from AlphaGrep’s alternatives business. This is an attributed trade-media interview, not a technical disclosure or independent audit; it does not identify model families, training corpora, data rights, deployment stage, agent permissions, or AI-attributed performance. The regional capture note records these routes together with the already surfaced Lingjun first-party Chinese strategy and conference record.

QuantSpeak: Renee Yao on explainable ML and the “third-generation” quant model

The full QuantSpeak archive contains 27 finance-native episodes, including several that a hedge-fund/AI title search can miss. The newly recovered March 16, 2023 episode with Renee Yao is mirrored by the CQF Institute YouTube recording. The episode identifies Yao as Neo Ivy’s founder and discusses explainable ML in portfolio management.

The captions add a distinct operating-model disclosure. At 05:26–06:41, Yao contrasts human-led historical-pattern research with training an AI system intended to generate new research ideas. At 16:36–19:26, she describes nonlinear interactions as the reason factor-by-factor explanations are difficult and says high-level understanding of useful information is the practical boundary she accepts. At 21:25–22:21, she says Neo Ivy built a system from scratch because existing quantitative-manager infrastructure did not support the approach. At 29:31–31:58, she says AI can automate parts of idea generation and may change the team and scale requirements of a quantitative business.

These are dated founder statements, not an independently verified model inventory. The episode does not disclose model weights, training data, feature list, explainability method, access controls, current staffing, production stage, portfolio authority, or performance. Automated captions contain recognition errors; the capture note preserves the raw transcript and source boundaries. The adjacent Neo Ivy first-party page separately describes machine learning, natural-language understanding, and high-performance parallel computing, but does not independently validate the interview’s operating claims.

The same archive yields three additional transcript-backed routes. Tony Guida’s RAM AI episode, mirrored on YouTube, identifies him as Co-Head of Systematic Macro and discusses model comparison using price-based features, alternative data, reproducibility, p-hacking, peer review, and out-of-sample discipline (00:49–02:53, 04:21–10:03). This is a named systematic-manager research route; it does not disclose RAM AI’s live model inventory or results.

Sonam Srivastava’s Wright Research episode, mirrored on YouTube, identifies her as founder and discusses dynamic stock/ETF allocation, reinforcement-learning feedback, LSTM/convolutional/recurrent architectures, and objectives that include return, risk, diversification, and churn (11:17–16:07). The episode is a dated founder-research account; it does not establish current production status, model ownership, or independently audited performance.

Samit Ahlawat’s J.P. Morgan episode, mirrored on YouTube, is a useful control comparator. Ahlawat discusses point-in-time data, leakage from same-day or future accounting data, survivorship bias, hyperparameter snooping, training/validation/test separation, parsimonious models, and a pre-specified path from calibration through deployment and recalibration (01:22–10:26). He also mentions synthetic-data research for finance (23:54–26:01); that is a research direction, not a deployed system. The combined QuantSpeak source note preserves the three transcript captures and their boundaries.

QuantSpeak: Grant Fuller on Irithmics and vicarious risk

The title-blind QuantSpeak archive also surfaces a vendor/data-modality route that a firm-name search would miss. In the 2023 episode with Grant Fuller, the CQF Institute recording identifies Fuller as co-founder and CEO of Irithmics. He describes “vicarious risk” as risk that other market participants see but a portfolio manager does not (08:01–09:16), and connects it to inferring institutional views from information and allocations (11:31–12:52).

Fuller says Irithmics calculates the measure for 177,000 listed companies globally and distinguishes the product from a signal or alpha generator, presenting it as information for portfolio and risk managers (15:08–15:41). Irithmics’ first-party product page separately describes deep neural networks, reinforcement learning, investor-behavior patterns, forecasts of exposure changes, and web/API delivery. The episode also names University of St Andrews relationships and student work on raw data and deep learning (05:06–05:45). This is vendor/CEO testimony plus product positioning. It does not identify a hedge-fund customer, proprietary dataset, data license, model specification, predictive performance, or live portfolio authority. The source note records the recovery and evidence boundaries.

QuantSpeak’s early AI/ML archive: infrastructure, pricing, hedging, and research practice

The archive also contains four previously unrepresented episodes that widen the evidence map beyond named managers. NVIDIA’s Tim Wood and John Ashley discuss GPUs, data, derivatives pricing, risk, and possible uses for text, audio, video, news, and regulatory filings (02:39–07:23; 40:58–42:00). This is vendor framing, not evidence of a customer implementation.

Arthur Böök’s episode describes a research workflow using S&P 500 option prices, neural-network interpolation of Monte Carlo prices, and explicit concern about rare-event data and interpretability (02:53–06:26; 08:09–09:42). The associated “Smiling in Action” research record makes this a technical research route, not a disclosed live fund system.

Jörg Kienitz’s Acadia interview adds model-agnostic, data-driven delta hedging, local regression, Gaussian-mixture models, open-source software, and C++ implementation themes (03:01–10:58). The UCT publication list corroborates a public research lineage. It does not identify a client, implementation, performance record, or trading authority.

Paul Wilmott’s 2023 round-up is practitioner commentary on AI-assisted coding and the role of mathematics in quant work (05:26–06:04; 25:26–26:07). It is useful for tracking public discourse, but it does not show an employer’s automation policy or deployment.

The combined early-archive source note keeps vendor, research, professional, and practitioner evidence separate.

G-Research’s GR-OSS OUT Podcast: title-blind engineering and AI infrastructure

The firm-controlled GR-OSS OUT Podcast was not visible in the seed queue as a conventional hedge-fund or quant podcast. Its JSON feed and RSS feed list 26 episodes, public MP3 enclosures, and linked YouTube videos. The archive is an open-source engineering and recruiting surface, so it should not be read as a disclosure of G-Research investment systems.

The first captioned priority routes are useful for infrastructure mapping. Building AI with Databricks, with Ben Wilson, discusses Unity Catalog AI’s agentic tool-calling framework, hosted foundation models, and ML/GenAI training cycles (01:08–01:16; 06:48–06:53). Making Kubernetes Work for AI and Batch Workloads, with Kevin Hannon, connects AI/ML workload support to batch scheduling, Armada, and Kubernetes working-group activity (00:00–00:16; 01:12–01:32). Imposter Syndrome contains an external practitioner account of using AI-generated code only when it can be independently explained, while OpenStack Ironic covers server-lifecycle automation, runbooks, and security guarantees. The linked YouTube captions are automatic and were used for timestamp navigation, not exact quotation.

The source adds a firm-controlled podcast, feed, audio, and video distribution layer to G-Research’s existing careers, code-review, open-source, and ML-infrastructure evidence. It adds no public finance-specific LLM, investment-agent permission, or performance evidence. The capture note records the episode map, routes, timestamps, hashes, and boundaries.

The title-blind Data Flowcast episode with Christos Bisias adds a separate G-Research engineering route. Astronomer identifies Bisias as an Open Source Software Engineer working with Apache Airflow at G-Research and describes large-scale data transformations, upstream contribution, OpenTelemetry tracing, CI-enforced metrics governance, and scheduler-throughput work. The publisher’s timestamps connect G-Research’s machine-learning/big-data market-prediction context at 01:20 with observability and platform details at 04:30–12:10; the Airflow Summit speaker page independently corroborates his G-Research Open Source role. This is engineering infrastructure and personnel evidence, not proof of a finance model, investment-agent authority, or performance. The capture note preserves the episode date and source boundaries.

QuantSpeak: named managers, volatility risk, and quant infrastructure

The archive also recovers four named practitioner routes that do not depend on an AI keyword in the episode title. Jan Rosenzweig’s Pine Tree interview discusses the tension between Sharpe-ratio allocation, equal-risk contribution, higher moments, and tail-risk-aware portfolio construction (00:55–04:48). Hari P. Krishnan’s SCT Capital interview discusses credit, positioning, carry, and agent-based modeling as lenses on abrupt volatility (01:54–06:19). These are risk and portfolio-methodology disclosures, not AI-system disclosures.

Misha Fomytskyi’s Vola Dynamics interview describes a path from physics and numerical coding through derivatives infrastructure, hedge-fund portfolio management, high-frequency market making, and a software company building curves and volatility surfaces (03:00–06:20). Elie Ayache’s ITO 33 interview adds a financial-software route centered on convertible-bond pricing, equity-to-credit modeling, and volatility-surface calibration (00:40–01:30). These sources map quant infrastructure and model-risk lineages; they do not establish AI deployment, customer use, performance, or capital authority. The source note preserves the dated affiliations and caption limitations.

QuantSpeak archive-completion screen: governance, emerging compute, and research lineage

The remaining 14 archive episodes were recovered and screened rather than omitted because their titles lacked hedge-fund or AI terms. Natalie Packham’s London Whale episode discusses correlation stress testing for stock and credit portfolios and the need for risk management to retain authority to intervene (00:48–01:18; 09:18–10:30). It is a model-governance route, not an AI deployment disclosure.

Araceli Venegas-Gomez’s quantum-finance episode describes quantum-computing proof-of-concept work, workforce training, and the need to test for a problem-specific quantum advantage (02:19–05:30). This is an adjacent emerging-compute and skills signal; it does not establish a production finance system.

Carol Alexander’s 2024 episode provides a public lineage across market-risk modeling, Algorithmics, high-frequency data, and academic research (00:54–01:55). She describes early model-building and coding work (06:19–07:19) and mentions a Journal of Banking & Finance special issue on GenAI and finance (11:49–12:54). This is researcher/community evidence, not an employer’s GenAI strategy.

John Guerard’s episode adds historical quantitative-research affiliations with Advantage Global Advisors and McKinley Capital Management (01:54–02:31). The archive-completion note records the remaining nine episodes as reviewed coverage without converting ESG, diversification, capital-valuation, historical-model, or conceptual quantum discussions into AI or current-firm claims.

New title-blind route — Hedgineer S3E16

The current Hedgineer episode “Can AI Turn You Into a Creative?”, dated August 18, 2026, is a title-blind vendor-media route. Its publisher description names an internal production workflow that uses Claude Code for storyboarding, Runway, Veo, Kling, and Seedance for video generation, ElevenLabs for narration and music, and Remotion for assembly. The description says the workflow was built in two days and then ran end to end from a phone. It explicitly asks whether the same pattern could reach investor relations, business development, and research.

This is useful evidence about a vendor’s own workflow and a search route that a hedge-fund/AI title filter would miss. It is not evidence of a named fund’s adoption, customer data access, model ownership, research deployment, permissions, investment authority, or performance. The capture note records the publisher cross-check and the absence of a promoted transcript in this pass.

LSEG fixed-income and regional title-blind routes — August 26, 2026

The fixed-income episode identifies David Rickard of LSEG and Roderick Joniaux of Tradeweb. Joniaux describes the growing role of hedge funds on Tradeweb’s government-bond platform and buy-side automation in the episode’s stated context (00:10:31–00:11:42). Rickard describes real-time and historical time series used for trading expectations, book marking, and risk-limit monitoring (00:12:45–00:14:27). Joniaux says hedge funds generally want raw data for their own analysis and discusses reference and pre-trade data (00:25:49–00:27:35). This is platform and data-readiness evidence, not a named fund’s model or live AI deployment.

The Middle East episode identifies Chude Chidi-Ofong of Montague Square Advisors, Rob Thomas of Centralis Group, and Jim Backhouse of LSEG (00:00:11–00:02:15). It distinguishes manager location from fund domicile and discusses Cayman, Dubai, Abu Dhabi, regulation, and operating infrastructure (00:02:15–00:14:37). It contains no attributable AI, ML, or GenAI disclosure, so it remains regional coverage rather than an AI finding. The capture note records both decisions.

LSEG title-blind AI and performance-data routes — August 26, 2026

The archive enumeration exposed three more episodes that a search limited to AI, hedge-fund, or quant titles would miss. The FX Execution episode identifies Tradefeedr’s Tim Cartledge, Bank of America’s Tan Phull, and Norges Bank Investment Management’s Alan Martin Lucero. Cartledge describes a possible language-model interface over “hardcore stats,” while warning against sending raw data to a language model and assuming the answer is correct (00:17:39–00:18:47). Lucero describes human execution and explainability around regulated decisions (00:18:50–00:20:06); Phull describes internal LLM proofs of concept and summarization use cases (00:20:21–00:21:00). These statements are not evidence of a named live trading model or autonomous execution.

The psychology and personalities episode identifies MarketPsych CEO Richard Peterson and performance coach Alden Cass. Peterson describes an AI analyst for deep-research reports, LLM use for data mapping and outlier detection, and experiments using candlestick-chart images; he distinguishes those experiments from statistical or time-series forecasting and says the discussion does not describe AI running portfolios (00:32:08–00:34:20). Earlier, Peterson describes MarketPsych’s news, social-media, and transcript sentiment feeds (00:12:18–00:13:54). These are vendor-reported product and workflow claims; the episode does not provide an independent dataset, model, validation, customer, or performance audit.

The performance-data episode identifies Essentia Analytics founder and CEO Clare Flynn Levy and Peterson. Levy describes decision-level analysis of daily holdings and trades as “investment episodes,” decomposing selection, entry timing, position changes, and exits, then comparing hit rate with winner-to-loser payoff (07:38–12:43). She also describes the use of this analysis in capital-allocation and pod-style risk conversations (19:24–22:08). This is a named vendor methodology account, not proof of a particular customer’s deployment or a validated AI system. The capture note records the source boundaries and registry decisions.

The discovery rule is now explicit: enumerate every episode by season and inspect its transcript, including execution, psychology, and performance-data categories. These routes expose interface design, image and language-model experiments, sentiment data, and decision-level analytics that keyword-only searches omit.

Independent publisher recovery: Bridgeway, Blackstone BXMA, and a registry gap — August 26, 2026

The Excess Returns interview with Elena Khoziaeva is a newly registered Bridgeway route found through a factor-investing title rather than an AI query. Apple and the publisher transcript route identify Khoziaeva as Bridgeway’s Co-Chief Investment Officer and Portfolio Manager and list “How Bridgeway is using AI” at 55:02. The transcript frames AI as an aid for data, text analysis, and trading-related research while preserving human judgment. Earlier timestamps describe replication, regime checks, statistical defenses against data mining, and team criticism before a research result becomes a model candidate. This is practitioner discussion about process; it does not disclose a named model, training corpus, data license, permission boundary, production status, or performance result. The capture note records the source hierarchy.

The J.P. Morgan Making Sense episode on Blackstone’s hedge-fund investing platform is another title-blind recovery. The official page identifies Joe Dowling, global head of Blackstone’s Multi-Asset Investing business, and Riad Abrahams, head of Strategy, Risk and Quant Analytics in BXMA. The accompanying publisher page and timestamped transcript route discuss BXMA’s manager selection, managed accounts, seeding, risk, and the use of data and AI tools to accelerate research, portfolio-company monitoring, and due diligence. Abrahams presents the breadth of Blackstone’s platform as a source of proprietary data and infrastructure; that remains a firm-reported description rather than an independently measured inventory. The episode is allocator/platform evidence, not a disclosure of a Blackstone manager’s model architecture, data contract, agent permissions, investment authority, or returns.

The March 2026 Quant / Financial Engineering episode was not a new capture: its audio and transcript already existed locally under the Quant / Financial Engineering source directory, but the episode was missing from the central media registry. The reconciliation adds the SoundCloud, Spotify, and Apple routes to coverage tracking. Its practitioner examples—earnings review, spreadsheets, Jupyter/Python, options-data inspection, and code/debugging—remain workflow vocabulary, not evidence of a large-fund production system.

The same registry-gap audit recovered the existing Invest with AI episode with Doug Clinton, with matching Apple, publisher transcript, and Podscan routes. The episode identifies Clinton as Intelligent Alpha’s founder and describes frontier models performing investment analysis and portfolio-management experiments, a hybrid quant/fundamental process, multiple data buckets, model routing, knowledge graphs, and explicit limits around human intuition and context. It is named practitioner evidence, not a complete model inventory, data-rights disclosure, permission map, unrestricted capital-authority record, or independently audited performance result. The existing source note is now linked to central coverage.

This pass changes the discovery rule in a concrete way: factor, allocator, and portfolio-construction titles must be searched for firm names, guests, transcript terms, and publisher cross-links. A title-level AI filter would have missed both Bridgeway’s AI segment and Blackstone’s data/AI discussion. The registry now records the difference between a genuinely new source and an existing local capture that had not yet been reconciled.

New UK personnel and research-lineage route — August 26, 2026

The title-blind UK pass recovered the March 24, 2025 Investology episode with Dr. Richard Saldanha. The listing gives the episode a 30-minute runtime and chapters on AI definitions, language models, limitations, AI in finance, financial data, computing, and quantum computing. The Institute of Science & Technology episode note describes Saldanha as an ex-hedge-fund manager and AI/ML practitioner and links a public AI-webinar playlist; those are publisher and institutional descriptions, not a named employment record. The matching Investology YouTube mirror supplies a public English automatic-caption track. The public Substack audio enclosure was also recovered and processed with local large-v3-turbo ASR. The reconciled caption/audio route supports timestamped navigation: Saldanha explains LLMs through statistical modeling, data, optimization, transformers, and attention (03:59–07:28); emphasizes targeted corpora, guardrails, and small language models for regulated finance (17:00–19:21); discusses order-book, tick, and auction data as research inputs (20:47–22:13); and treats hedge-fund-funded foundation-model work as a separately financed technology project, not a generic attribute of all hedge funds (22:31–25:47). These are bounded paraphrases from automatic/non-diarized transcripts and speaker-reported conceptual commentary, not evidence of a current Oxquant or fund implementation. See the capture note.

The Queen Mary University of London profile provides a direct personnel and lineage record: Saldanha is listed as a Visiting Lecturer, with expertise in statistical machine learning and quantitative finance, more than two decades of risk, trading, and fund-management activity at unnamed City of London institutions, and current engagement through Oxquant. The same profile identifies an Oxford DPhil in graph theory and multivariate statistics, involvement in the Alan Turing Institute’s AI for Control Problems project, and public work spanning large language models, reinforcement-learning competition infrastructure, and AI/finance education. This makes the route useful for academic-lineage and research-topic mapping, while leaving former employers, current fund affiliation, model ownership, data rights, investment authority, and performance unresolved. The capture note records the evidence hierarchy and follow-up queue.

Firm-controlled lab and research surfaces — August 26, 2026

The latest firm-page refresh adds several non-podcast routes to the discovery graph. Bridgewater’s current partnership page names Blake Cecil as Deputy Chief Investment Officer for Alpha Engine and AIA Labs, Nina Lozinski as Head of AIA, Theo Saarinen as Chief Research Officer for AIA Labs, and Oliver Simon as Head of AI & ML Investment Strategy. Its AIA research page separately names Rohan Alur, Daniel Kang, and Yuxuan Zhu and describes RLVR experiments with Qwen3.5-4B across math, programming, general-knowledge reasoning, and Text-to-SQL. These are first-party titles and research artifacts; the experiments are not investment benchmarks and do not disclose a production investment model, proprietary financial training data, agent permissions, or return attribution. An older Bridgewater people-page title for Jasjeet Sekhon conflicts with his current profile, so the discrepancy remains date-scoped rather than resolved by inference. The capture note preserves the conflict.

The August 27 Daniel Kang X thread and the linked Thinking Machines report add a distinct, title-blind research route. They name Bridgewater AIA Labs, UIUC, and Thinking Machines; describe Tinker-based RLVR fine-tuning of Kimi-K2.6 on the corrected BIRD-Platinum data; and specify reward checks for semantic equivalence and supplied domain knowledge. The report gives benchmark-specific accuracy and cost figures for Arcwise-Plat-SQL, but the public artifact remains a text-to-SQL research result. It does not establish a PAT implementation, an investment workflow, a live production model, or an AI-attributed return. See the timestamped source note.

Jump’s AI/ML page states that AI and machine learning span trading, research, and infrastructure, including classical models, deep learning, reinforcement learning, LLMs, generative modeling, real-time inference, NLP, and centrally built foundation models and agents. A separate Vera Rubin announcement names Joe Stam as Head of Research Technology and Alex Davies as Chief Technology Officer and describes a financial-research data-center expansion. Jump’s fellowship page adds an academic-network route with AI/ML fellows from Harvard, MIT, and Stanford. The pages are self-described operating, infrastructure, and talent surfaces; they do not publish model weights, training corpora, vendor contracts, evaluation protocols, agent permissions, capital-allocation authority, or independently audited performance. The capture note keeps those evidence layers separate.

Renaissance Technologies provides a useful negative control. Its official About page and current research roles describe scientific hiring, data processing, GPU/CPU infrastructure, statistical models, and systematic trading. The reviewed pages do not name an AI lab, LLM, generative model, agent, or current AI owner. This supports a statement about what the public pages expose—not a conclusion about what the firm does privately. The Renaissance refresh note records the page-level disqualification loop.

Ubiquant’s English company page, UbiquantAI GitHub organization, and public repositories add a Chinese AI-lab and model-engineering route. The public artifacts include iDO, a local desktop assistant; Qwen2.5-Math-7B post-training research; a Universal Reasoning Model; and Fleming medical reasoning and multimodal models. Zitian Gao’s public profile identifies him as a researcher at Ubiquant and links to two repositories. These are public company, code, and personnel signals across productivity, reasoning, and medical modalities; they are not evidence of a financial model, trading permission, or transfer of those artifacts into an investment system. The Ubiquant capture note keeps repository authorship separate from a complete employee roster.

G-Research’s quantitative-research and machine-learning page and GenAI engineering profile expose a current research-lab and personnel surface. The pages describe mathematical and statistical research, large datasets, hypothesis testing, a Machine Learning College, and an AI Engineer within a GenAI Engineering team delivering capabilities across the business. G-Research’s ELLIS partnership and university scholarship pages add a talent-network route. The profile does not publish a surname, model inventory, finance-specific assignment, permission map, or investment result; the academic partnerships do not establish employment or exclusive research access. The G-Research source note records the named-but-partial personnel boundary.

Fresh title-blind allocator and former-quant routes — August 26, 2026

The Investment Researcher’s Podcast episode from Zenith Investment Partners was not in the registry. Its Apple and Amazon listings identify Zenith analysts Ethan Spiegel and Bradley Antman and describe an Australian allocator/research-provider discussion drawing on manager conversations. The published description names drafting, document summarisation, quantitative signal testing, and due-diligence questions about tooling, proprietary insight, and team composition. Zenith’s research-team material independently anchors Spiegel’s current analyst role and education. This is an allocator-side discovery and diligence route, not a disclosure of any manager’s model, data rights, production status, permissions, authority, or performance; the episode’s coverage counts remain publisher-reported until the audio or transcript is recovered. See the capture note.

The Personable episode with Joseph Mezrich adds a former quant-leadership route found through title-blind podcast search. The listing and Mezrich’s public post describe his former Nomura, UBS, Morgan Stanley, and Salomon Brothers roles, current Metafoura activity, and a discussion of model limits, uncertainty, human understanding, and AI safety. The CFA Institute record of his 1994 paper supplies a separate publication anchor on decision trees and linear investment models, while an archived Nomura biography preserves the public education and historical-ML description. The public RSS enclosure was recovered and processed with temporary local English large-v3 ASR. Mezrich frames AI as an answer-producing tool whose outputs still require human understanding (14:42–19:40), says models eventually break and discusses the need for supervision/kill-switch thinking (21:20–24:50), recalls publishing machine learning for option hedging at Salomon Brothers in 1994 (26:57–27:12), and argues for resilience and diversification because black-swan events cannot be forecast (29:43–31:05). This remains historical personnel and methodology evidence; nothing here establishes a current hedge-fund system, model inventory, permissions, authority, or result. See the capture note.

The recovered Alternative Data Podcast episode with Evan Reich adds a full-episode route behind the earlier System2 LinkedIn clip. The episode describes data sourcing as an interface between quant, fundamental, research, legal, and compliance teams (05:39–05:55), says AI-enabled data-loading tools could reduce some loading work while leaving human interaction and creativity important (15:27–15:55), and discusses licensing limits on loading vendor data into an LLM (56:06–57:05). It also supplies a temporal personnel lead: the host says Reich has left Verition and is now Chief Product Officer and Head of AI at BWG Global; Reich then describes BWG as combining BWG and Off the Record research operations and exploring AI-mediated research consumption (53:20–55:43). This is host/guest-reported and requires first-party BWG confirmation. It should not be back-projected onto Verition, Millennium, SAC, or Quandl. See the ASR recovery note.

The recovered Charles-Albert Lehalle episode adds a historical CFM/ADIA route that was previously metadata-only. Lehalle describes creating data-analytics teams during his final three years at CFM, satellite imagery as an early project, and a similar approach at ADIA (15:54–17:03). He describes turning high-dimensional alternative data into lower-dimensional inputs for standard alpha teams (17:03–18:27), and discusses data lineage and pipeline instrumentation for diagnosing performance changes (27:28–31:54). A separate conceptual section discusses causality, world-model ideas, and embedding-space representations (12:09–13:48; 45:13–47:35). These are speaker-reported historical and conceptual claims, not evidence of current CFM/ADIA systems, vendors, permissions, or performance. See the ASR recovery note.

The recovered Christina Qi episode adds a market-data-vendor route tied to former Domeyard HFT experience. Qi says Domeyard used machine learning but that some of its best-performing strategies were comparatively simple (13:40–15:57). As Databento’s founder, she says unnamed AI firms bought market data for direct AI use cases and may be training on it (32:42–33:07); no customer is named or confirmed (33:24–34:10). She distinguishes conventional market-data ML from LLM-assisted analysis, including users connecting an API key to an LLM or coding assistant (35:50–37:50). These are speaker-reported vendor observations, not evidence of a Databento partnership, customer model, training right, or hedge-fund deployment. See the ASR recovery note.

The French-language pass recovered an IRIVEST first-party page for Blindé with Stéphane Levy, dated November 24, 2025. IRIVEST identifies Levy as Strategist and Head of Innovation and says its process combines price momentum, earnings momentum based on consensus analysis, and an AI stock-picking model introduced in 2019. The linked YouTube recording is now captured as a French automatic-caption transcript: at 13:18–14:54, Levy describes a 1,500-share European universe and two monthly rebalancings; at 33:19–34:14, he describes a neural-network component whose stated inputs include historical performance, price momentum, and earnings momentum; and at 37:32–37:58, he frames the model as a shared team asset. These are speaker statements in unverified ASR, not a model card or independent audit. No architecture, retraining schedule, validation design, permissions, deployment telemetry, or AI-attributed performance is established. The capture note preserves the French wording boundary.

New first-party EQR video — August 27, 2026

The title-blind topical sweep recovered a short Citadel video featuring Perry Vais, identified on screen as Head of Equity Quant Research. In the firm-controlled recording, Vais describes equity quantitative research as forecasting company drivers from observable data artifacts (00:15–00:35), the intersection of quantitative research with large-scale systems and data analysis (00:40–00:48), and an opportunity-first sequence of assembling a team, determining models, and sourcing data (02:37–02:51). He also describes cross-team technical support and iterative researcher development (01:15–02:04). This is first-party evidence about a named EQR role and operating description. It does not identify a specific AI or generative-AI model, training data, validation protocol, production permission, order or portfolio authority, or independently measured result. The capture note records the transcript hash and the EQR/Citadel Securities entity boundary.

A fresh subtitle recovery from the Citadel Securities official YouTube archive upgrades seven previously metadata-only records to timestamped caption evidence. In the Peng Zhao–Alfred Lin Future of Markets discussion, the panel distinguishes machine-learning models from LLMs and agents (04:46–05:17), describes unique data, data-use insight, algorithm-capable people, and workflow as four ingredients in AI capability (13:52–14:28), and discusses coding agents, search, summarization, compliance forms, and the choice between automating a workflow and waiting for tools to improve (32:45–35:10). A Jim Esposito item links AI investment to operating-cost reduction, client experience, data curation, risk-model revision, and speed to market (00:00–01:34). Title-blind recruiting shorts for Yiming and Sriram add statistically tested signal research and systematic-equity-options quoting workflows (00:00–01:21). These are first-party captions and recruiting narratives, not speaker-adjudicated transcripts; they do not disclose model inventories, data rights, permission boundaries, autonomous trading authority, or performance. The archive capture note records caption hashes and proper-name caveats.

Bridgewater PAT recording recovery — August 27, 2026

The queue audit found and recovered the previously unreviewed LangChain recording of Bridgewater’s PAT presentation, published July 24, 2026. It is the same Interrupt 2026 case study represented by Bridgewater’s AIA Labs page, so it is reconciled to that record rather than counted as a second deployment. The recording names Brendan McManus, Michael Ran, and Santi Weight as presenters. McManus describes PAT as an exploratory-investigation tool rather than a trading system (04:29–04:43), and describes a broader research-circle design in which discrete sub-agents support perception, question formation, investigation, synthesis, and updating shared understanding (03:11–04:13).

The recovered captions add operational detail to the existing record. They describe per-user context and tool configurations aligned to information entitlements (09:29–10:20); search over broker research, earnings transcripts, internal emails, and other documents alongside structured and internally derived time series (10:24–11:29); and a plan that specifies data frames, schemas, and dependencies before parallel sub-agents generate Python/Pandas analysis (12:49–13:45). The talk also describes background review of completed interactions, human-audited benchmark creation, regression checks, and a user-facing “Teach” flow that proposes context or harness changes through a pull request (06:24–06:52; 15:18–16:41). Weight identifies LangGraph for the investment-context chat agent and separates it from a coding layer presented as ordinary Python with static analysis, dependency validation, and cached execution (18:01–20:18; 20:04–24:19). These are presenter-reported architecture and workflow details. The recording does not establish model weights, training-data rights, evaluation denominators, current employment beyond the recording, per-agent capital authority, or independently measured investment performance. See the capture note.

Man Group / Oxford-Man — DeepLOB and execution research recovered from a title-blind archive route

The Risk.net Quantcast episode with Stefan Zohren, linked to a recoverable SoundCloud recording, was absent from the earlier registry. Risk.net identifies Zohren as an Oxford-Man Institute associate professor and a Man Group researcher responsible for futures and foreign-exchange execution research. The episode is a useful title-blind route: its title names DeepLOB and forecasting rather than hedge funds or AI.

The recording describes the model lineage in concrete terms. Zohren explains DeepLOB as a convolutional architecture applied to space-time limit-order-book representations, with local features such as imbalance and temporal components; the multi-horizon extension uses sequence-to-sequence and attention mechanisms to generate a forecasting path (05:00–15:00). The underlying papers are available from arXiv for DeepLOB and arXiv for the multi-horizon extension. This is market-microstructure modeling, not a text or language-model system.

The firm-specific disclosure is bounded but material. When asked whether the work was used in financial institutions, the interviewer names Man Group. Zohren confirms that the earlier DeepLOB work had been used as the basis for work on Man Group’s in-house algorithms, while saying the newer multi-horizon model had not yet been implemented as a specific model at the time of the interview (15:40–17:50). The discussion frames the signal as useful for execution and market making, including a possible choice between crossing the spread and maintaining a passive order (15:58–17:05). The recording does not identify a current production model, exchange coverage, model version, deployment permission, or performance record.

The episode also exposes adjacent research directions. Zohren describes reinforcement learning for execution using execution cost or slippage as the reward, order choice as the action, and forecasts as part of the market state (18:08–20:50). He discusses IPU-based hardware experiments and reports a five-to-ten-times speed-up in that dated setup relative to the GPU comparison (20:50–22:50). Later, he describes machine-learning-enhanced risk and momentum work, learning-to-rank for cross-sectional strategies, transformer-based time-series momentum, regime changes between momentum and reversion, and attention-based interpretability (23:00–28:55). These are research and practitioner statements from a 2021 recording, not evidence of current strategy composition or returns. The timestamped capture note records the local audio, ASR hashes, and evidence boundaries.

Ritter Alpha — execution cadence and reinforcement-learning research

The Risk.net Quantcast episode with Gordon Ritter and its recoverable SoundCloud recording add a second title-blind Risk.net route. Risk.net identifies Ritter as CIO and founder of New York-based Ritter Alpha and Jerome Benveniste as a senior research scientist and co-head of equity statistical arbitrage. The episode also links the underlying optimal-turnover paper.

Ritter explains an execution framework that relates turnover to signal autocorrelation, risk aversion, liquidity, and volatility under a Gaussian-process and linear-price-impact setup. He describes it as a rule of thumb for checking whether a strategy’s turnover is in a plausible range and says the fund-era implementation was updated more frequently than hourly (14:15–17:42). He also notes that real implementations must account for bid-offer spreads, intraday cost structure, and the difference between continuous-time formulas and discrete trading (16:36–17:42). This is a dated speaker account, not a current execution-system inventory.

The reinforcement-learning discussion is similarly specific but bounded. Ritter frames optimal execution, alpha-decay trading paths, derivative hedging, and dynamic replication as multi-period stochastic-control problems. He describes reward design, deep-neural-network policy methods including PPO, and the use of RL as a numerical route when a closed-form control solution is impractical (24:55–36:20). He discusses derivative-portfolio and hedging extensions as ongoing research (37:25–42:00). The audio does not establish a deployed RL policy, model weights, training data, permissions, capital authority, or independently measured performance. See the timestamped capture note.

Man Group / Oxford-Man — corrected historical and market-impact archive routes

Two additional Man/Oxford-Man records extend the timeline beyond the recovered podcast interviews. A 2016 Risk.net article describes Man AHL’s use of machine-learning algorithms associated with its Oxford research unit. The same page carries a September 9, 2016 correction: preliminary testing used seed capital earmarked for researching new algorithms rather than shareholders’ money. The accessible preview truncates the article’s subsequent performance language, so this record is evidence of a dated research-and-seeding process, not an audited return result or a basis for repeating the headline as performance evidence.

A 2023 Risk.net article reports that an Oxford-Man team built a machine-learning model intended to forecast how large trades might move prices. Its public opening describes an LLM-type system that identifies patterns in messages sent to a limit order book. The article is access-limited after its opening paragraphs; it does not publicly establish a model name, architecture, training corpus, evaluation design, exchange coverage, current deployment, or production ownership. Man Group’s 2022 first-party OMI announcement separately says OMI research had affected client programmes in active risk overlays, trade execution and order-routing, and transaction-cost and market-impact monitoring. That is a firm-reported application statement, not a mapping from any one paper or model to a live strategy.

These records materially widen the public evidence surface around execution and market-impact research, while preserving the key disqualifications: no current model inventory, proprietary data rights, agent permissions, capital authority, or independently audited AI-attributed performance is disclosed. The capture note records the source hierarchy, access limits, and follow-up route.

Duality and Qognitive — forecasting, sparse-data similarity, and partner routes

The title-blind Risk.net archive adds a 2020 interview with Dario Villani, identified by the publisher as co-founder and CEO of New York-based Duality Group. Risk.net describes Duality as using machine-learning-led algorithms across US stocks, ETFs, and global futures. In the recovered SoundCloud recording, Villani describes a few-day alpha horizon (02:15–02:43), a neural-network forecasting core (13:25–14:49), and additional ML use in feature engineering, execution-quality and broker/venue selection, and corporate-action NLP (13:53–14:49). He discusses saliency-map attribution for an ensemble of competing models (19:01–20:13), rare-event handling through ensemble weighting (30:42–36:22), and mean-field/agent-interaction ideas for execution and market impact (38:06–39:05). He also describes a few-dozen-person firm with offices in Kyiv, Warsaw, and New York at the time of the interview (42:11–42:50). These are date-scoped practitioner statements, not a current roster, model card, permission record, or independent performance study.

An additional Real Vision interview with Jim Pallotta supplies a distinct 2019 title-blind route. Real Vision’s description links Pallotta with Raptor Group’s exploration of machine learning and AI; the opening identifies Dario Villani as Duality’s CEO and founder and asks Pallotta about his backing relationship (00:05–00:22). Pallotta describes a Duality asset-management activity alongside an analytics business being spun out and gives an example in which the analytics activity could screen approximately 40,000 datasets, identify about 1,500 as relevant for the current period, identify relevant time windows, and update that selection (01:27–02:20). This is a speaker account in a third-party interview, not a Duality technical specification or independent customer result. The capture note records the date, transcript route, and the unresolved surnames of two people Pallotta names only by first name.

The same archive adds a 2025 Villani–Musaelian episode with a recovered SoundCloud recording. Risk.net identifies Villani and Kharen Musaelian as co-founders of Duality and Qognitive, and describes Musaelian as Duality CIO and Qognitive chief science officer. The episode discusses Qognitive’s QCML approach for identifying tradable bond alternatives in sparse and outlier-heavy settings, with a paper developed with BlackRock collaborators (18:21–22:23). It also describes an equity similarity application and possible healthcare and robotics uses (34:20–46:33; 49:15–50:02). The speakers say they were using small language models for claims while avoiding large language models at that time because of expense and data requirements (49:28–50:02), and describe an IBM joint development agreement for quantum-hardware applications (27:16–27:55; 01:01:23–01:02:06). The partner references do not, by themselves, establish BlackRock, Deutsche Bank, or IBM deployment in a Duality investment book.

The deeper recovered audio also provides a dated architecture description: the speakers discuss QCML as a feature-rich representation and similarity method, distinguish quantum mathematics from execution on classical computers, and place the finance example in illiquid-bond substitution (09:14–14:13; 18:21–22:23; 25:13–27:04). Qognitive’s current team page lists Villani as co-founder and CEO and Musaelian as co-founder, president, and CSO; its bond paper supplies the formal method and evaluation context. The recording separately covers equity linkages, patient and insurance-claims similarity, healthcare product direction, and robotics (34:20–50:02), plus a speaker-reported IBM Heron experiment and joint-development route (01:01:16–01:04:56). Those are separate application and partner layers. The evidence still does not identify a current investment-book implementation, full model inventory, training-data rights, production permissions, agent authority, or independently audited results. See the expanded capture note.

The 2020 Horvath–Lee episode supplies an adjacent quantitative-finance research route. Risk.net identifies Blanka Horvath at King’s College London and Gordon Lee in UBS quantitative analytics, and describes variational autoencoders, signature-based path representations, synthetic price paths, and data augmentation for deep-hedging research. The publisher’s index places those topics at 02:15–33:30; the recording was not recovered in this pass. This is research-methodology evidence, not evidence of a hedge-fund system.

The timestamped capture note records the local ASR artifacts, source hierarchy, partner-entity boundaries, and unresolved questions. None of these records establishes current model ownership, proprietary training-data rights, validation denominators, live permissions, capital authority, or independently attributed returns.

Risk.net archive expansion — portfolio correlations, XAI, Man Numeric, and synthetic data

The title-blind archive pass also recovered a 2019 Risk.net multi-manager article that names Acadian Asset Management, XAI Asset Management, AllianceBernstein, Data Capital Management, Lazard Asset Management, Neuberger Berman, and Robeco in connection with machine-learning approaches to portfolio correlations, clustering, regime forecasting, factor exposure, and unstructured inputs. For Acadian, the article describes explainability and guardrails as important constraints around portfolio-allocation use. For XAI, it reports a shallow neural network in portfolio optimization with relatively small datasets. This is dated third-party reporting; it is not a current model inventory, deployment record, data-rights statement, or performance comparison.

The same archive supplies a January 2020 XAI profile. Risk.net identifies Aric Whitewood, formerly Credit Suisse’s head of data science, and Jonathan Wilmot, formerly its chief global strategist, as XAI co-founders. Its public opening reports that XAI’s core machine-learning system had been trading real money across several asset-allocation strategies for 18 months at publication and describes a proprietary macro business. The remainder is access-limited, so this record does not establish the system’s architecture, training data, current status, authority boundary, or independently measured performance.

A 2023 Risk.net account of Man Numeric’s SVB research adds a credit-data and network-analysis route. The public opening reports that Man Numeric attributed an early warning about Silicon Valley Bank to alternative data and network analysis of credit markets, and describes data automation and systematic processes applied to bond-market research. Risk.net described Man Numeric at the time as a $36 billion equity and credit manager. Those are dated, firm-attributed statements in third-party reporting; the page does not establish the exact data sources, graph construction, point-in-time controls, alert timing, model ownership, trade decisions, or independently verified results.

Finally, Risk.net’s 2021 synthetic-data coverage, with its companion commentary, provides a broader quantitative-finance modality route. The articles describe machine-learning-generated market paths as a way to test strategies against more than one historical realization, while warning that historical training may not capture regime changes, market microstructure, or participant motivations. This connects to the Horvath–Lee market-generator research above but is not evidence of a named manager’s production system. See the archive capture note.

Two additional Risk.net archive routes sharpen the distinction between signal research and control infrastructure. The 2019 QMA profile reports that an algorithm reading earnings-call transcripts was the firm’s best-performing stock-return indicator in the prior year. QMA CEO Andrew Dyson is quoted as describing AI testing at the margins, while retaining a boundary against using it as the decision-making tool. The article also describes NLP-based news scraping and earnings-transcript analysis for signals intended to be orthogonal to existing inputs and subject to ordinary evidential tests. This is dated third-party reporting and an executive account; it does not disclose the model, corpus, labels, point-in-time controls, current status, or independently audited result.

The 2019 AQR “Intercept” profile is a separate, non-AI-specific control-plane record. Risk.net identifies CRO Mike Patchen and describes an enterprise risk platform intended to show which risks facing AQR were elevated at a given time. The public page becomes access-limited after its opening. It therefore supports the existence and stated purpose of a dated platform, but not its data feeds, machine-learning components, thresholds, user permissions, escalation policy, current operation, or investment outcome. The archive capture note records both routes and keeps them separate from current firm deployment claims.

A distinct direct hedge-fund route is Risk.net’s 2024 profile of Goose Hollow Capital. Its public opening describes a boutique manager using generative-AI models to filter thousands of news sources for non-linear relationships. The article gives a December 2022 example in which the system surfaced a Russian-flagged ship docking in Johannesburg and the firm removed its South African exposure. This is a dated third-party report of a firm-attributed workflow and response; the public opening does not disclose the model, prompts, retrieval design, data rights, latency, human review, threshold, position sizing, false-positive rate, or realized result. The body is access-limited, so no additional detail is inferred. See the bounded capture note.

RAM Active Investments adds a current agentic-development route. Risk.net’s February 12, 2026 profile, independently repeated by WatersTechnology, identifies CIO and co-founder Emmanuel Hauptmann and reports that Hyperfold combines seven alpha-generating strategies, six created with assistance from agentic AI. The profile reports Hauptmann’s claim that development time fell by 50%–70% and describes an AI model factory and agentic credit analysts. RAM’s first-party research page separately lists LLM-based newsflow representations, fine-tuning LLMs for stock-return prediction, financial sentiment, and climate-controversy detection. The trade-press body is access-limited and the first-party page is a research-positioning surface: the public record does not disclose the six strategies, model families or weights, training-data rights, tools, evaluations, approval gates, portfolio permissions, or AI-attributed returns. See the source note.

Vinva — systematic scale, internal AI/ML, and earnings-call voice boundaries

The title-blind pass recovered an Inside the Rope interview with Morry Waked, Vinva Investment Management’s Managing Director and Head of Investments, with a public SoundCloud route. Waked describes a systematic “glass box” process in which data and technology scale analysis across thousands of companies while people provide investment ideas, model architecture, and judgment (22:45–25:08; 36:22–37:15). He says Vinva has its own GPUs, runs AI models internally rather than in the cloud at that point, and applies proprietary models to broad company relationships (30:16–33:32). The episode reports approximately $43bn at recording time; Vinva’s current home page reports $65bn+ AUM as at August 4, 2026, so the two figures are not treated as a single time series.

The episode is also a direct voice-modality discovery route. Asked about analyzing conviction in earnings-call voices, Waked says Vinva used conference-call information as far back as 2007–2008, but that transcripts, sentiment tools, and management coaching had reduced the distinctiveness of the signal (33:32–34:27). This is historical practitioner evidence and a public crowding observation, not proof of a current acoustic model, deception classifier, feature set, labels, timing, or AI-attributed performance. Vinva’s team page names research and technology personnel including Juliane Krug, Danny Lo, Therdsak Tangkuampien, Robert Franklin, Reshma Joseph, Ethan Hansen, Jenny Chen, Duncan Forrest, and Gangeyan Ganesalingam, with public quantitative, ML, data, and systems backgrounds. The distributor’s April 2026 quarterly update separately states that AI and ML support research productivity, data sourcing, and investment-signal development. See the capture note.

Goldman Sachs Marquee AI — institutional research-product and agent-control route

Goldman Sachs Exchanges’ August 24, 2026 episode identifies Chris Churchman as head of Marquee and co-chair of the Global Banking & Markets AI Working Group. Churchman describes Marquee AI as internally available at the time of recording: it decomposes user questions into research topics, retrieves firm research, trading-floor commentary, Market View analytics, and pre-trade tools, and can write and run Python for auditable calculations (02:00–05:14). He describes sentence-level grounding, source provenance, entitled access, secure agent environments, and a “mandate engineering” question about what an agent may do under whose authority (04:15–05:14; 09:50–11:22). The episode also distinguishes automation from reconceiving workflows around the real job to be done (18:15–20:31).

This is first-party asset-manager and capital-markets platform evidence, not hedge-fund deployment evidence. It does not identify the model provider or weights, training corpus, benchmark denominator, routing policy, client rollout, production permissions, trade authority, or AI-attributed performance. See the capture note.

New title-blind episode — Lucas Schuermann, Q Capital, Genesis, and Variational

The August 21, 2026 Odds on Open episode identifies Lucas Schuermann as founder and CEO of Variational and discusses his earlier Q Capital and Genesis Trading work. The public transcript describes historical market-neutral, algorithmic FX and crypto strategies, statistical-arbitrage-style relationships, inter-exchange arbitrage, and basis (00:00–02:18). It also describes a research path spanning differential geometry, computational neuroscience, machine learning, robotics, and Bayesian machine learning (05:32–06:27), followed by construction of an electronic market-making system at Genesis (08:10 onward). The publisher description adds on-chain perpetuals, aggregated liquidity, growth-curve data, and AI’s effect on technical skills. This is useful cross-platform and title-blind discovery evidence, but not evidence of a current GenAI system, model inventory, data rights, agent permissions, trading authority, or performance. The capture note records that the local YouTube caption request was blocked by bot verification; the transcript is therefore treated as publisher-hosted text rather than an audio-verified local transcript.

August 27 Shanghai regional-language conference routes

The Shanghai Advanced Institute of Finance notice schedules a September 4, 2026 forum on AI, compliance, and quantitative-investing talent. The organizer names SAIF practice professor Kan Rui, Zhongtai Securities technology executive Xu Hao, and Mingyue Fund partner and equity-strategy portfolio manager Zeng Lingqi; the program includes a session on AI applications in quantitative trading and a description of Zeng’s organizer-reported AI-assisted factor-research work. The Cailian Press report on a July 19 Lujiazui roundtable adds Chinese-language discussion of data validation, ontology and rule layers, factor and backtest generation, staged agent controls, and live-validation requirements. These pages add a regional conference and title-blind discovery route, not a recording, model inventory, permission map, production confirmation, or performance result. The capture note preserves the entity boundaries and translation caveat.

August 27 adjacent investment-firm AI leadership routes

B Capital’s appointment announcement names Dr. Andrew Jackson as General Partner and Chief AI Officer across investment, portfolio-management, and operations teams. Axios adds a publisher account of Bee Hive, an internal investing system described as preparing investment-committee material, taking notes, structuring deal arguments, and recording passed decisions for later review. B Capital is a venture-capital firm rather than a tracked hedge-fund manager, and the sources do not establish model versions, data rights, evaluation results, or final investment authority. Generali Investments separately announces Ole Jorgensen as Chief AI Officer across a multi-affiliate asset-management platform, with prior AI-in-investment-process work at Global Evolution Asset Management and two MIT strategic-AI programs. These are adjacent investment-manager leadership and workflow routes, not evidence of a tracked firm’s model deployment or investment performance. See the capture note.

New Australian title-blind route — Pendal / Elise McKay

Pendal’s June 12, 2026 first-party page identifies Elise McKay as an investment analyst and portfolio manager on its Australian equities team and links The Point podcast through Apple, Spotify, and Omny.fm. The Omny player exposed a 9:16 MP3, recovered and transcribed locally with WhisperX-MLX. At 00:39–02:51, McKay lays out five layers—power, chips, infrastructure, models, and applications—and places Australia’s opportunity primarily in infrastructure; at 03:52–04:29, she names Australian neocloud and infrastructure examples; and at 05:02–07:22, she describes software-screening criteria including product demand, AI enablement, execution, distribution, cost, and regulatory relationships. These are timestamped local-ASR paraphrases of an investment discussion, not evidence of Pendal’s internal AI tooling, model training, agent permissions, production deployment, or AI-attributed performance. The transcript was not diarized and is not a verbatim public transcript; see the capture note.

August 28 Curious Quant title-blind HFT and AI boundary route

The archive recovery added a title-blind Curious Quant episode with Christina Qi, normalized against her public professional site and LinkedIn profile. The episode describes a former Boston HFT operator’s division of labor between latency-sensitive trading and slower AI use: limited AI for strategy selection and obscure strategies, more plausible middle/back-office use, and open-source tooling for internships (17:37–19:37). Qi also describes prioritizing researchers and internal process improvement over paying to be the very fastest, and says strategy durability can range from weeks to years (20:38–22:15; 30:53–31:42).

This is useful as a boundary case for the wider map: “HFT” should not be treated as synonymous with broad AI deployment. The recording does not identify current model versions, data rights, production endpoints, agent permissions, or independently measured performance. It is practitioner and historical context, not evidence of any tracked firm’s current operating model. See the timestamped capture note.

August 28 Risk.net title-blind MIT adaptive-markets and ML-governance route

The unfiltered Risk.net archive recovered an Andrew Lo Quantcast episode whose title does not expose its machine-learning content. The direct SoundCloud recording and MIT Sloan profile resolve the transcript’s “Andrew Lowe” rendering to Andrew W. Lo, Harris Professor of Finance and director of MIT’s Laboratory for Financial Engineering.

Lo distinguishes algorithmic execution from delegating investment decisions to algorithms (19:43–21:04), relates the August 2007 quant unwind to common statistical-arbitrage exposures (21:50–23:32), and describes broader market-participant data as a potential input to dislocation analysis (24:22–26:37). He treats machine learning as a possible hypothesis-generation tool, while emphasizing prior worldview, rigorous testing, multiple-search correction, interpretability, and inductive-logic approaches (29:35–45:37). The episode’s future-work discussion combines behavioral science, AI, machine learning, data analysis, neuroscience, and psychology for portfolio and risk tools (46:11–47:15).

This is academic-practitioner methodology and title-blind discovery evidence, not a current hedge-fund deployment disclosure. It does not establish a tracked firm’s model inventory, training data, provider, production endpoint, permissions, or performance. See the timestamped capture note.

August 28 Risk.net JPMorgan/XTX deep-hedging and simulator route

The recovered Hans Bühler Quantcast episode provides a dated bank-research route with an executable SoundCloud recording and related drift-removal paper. Risk.net identifies Bühler as JPMorgan’s global head of equities analytics, automation and optimisation at the time; its later personnel report records his move to XTX Markets, and an XTX regulatory disclosure lists him as co-CEO in the reviewed document. The role sequence does not prove research transfer or XTX deployment.

Bühler describes statistical hedging for index vanilla products and expansion toward S&P and single-stock options, while placing deep hedging around complex products such as cliquet options (03:31–05:21). He explains drift removal as a guard against a neural network learning unintended directionality, and says the resulting trading strategy remains within established classic risk, capital, and limit frameworks (06:42–11:25). He also discusses GANs, variational autoencoders, signatures, market simulators, and training higher-order models on simulated paths when historical observations are limited (09:22–13:46). The episode records his study under Hans Föllmer and Alexander Schied (11:52–12:31).

This is a concrete, date-scoped bank implementation and personnel route, not proof of current XTX implementation or any tracked hedge fund’s use. The sources do not disclose complete model weights, data rights, evaluation denominators, autonomous permissions, or performance. See the timestamped capture note.

August 28 Risk.net/HedgeFacts nonlinear-risk and quant-research route

The recovered Matthew Dixon Quantcast episode and SoundCloud recording add a title-blind risk-technology route. Dixon’s public research site identifies his work across machine learning, quantitative finance, stochastic control, and financial model risk. At 29:34–30:30, he says the nonlinear risk-decomposition method was developed for HedgeFacts, described in the episode as a cloud risk-management solution for hedge funds.

The episode covers nonlinear delta-gamma risk decomposition across factors, managers, instruments, and subportfolios, with cross-factor correlations (00:18–01:19; 11:11–14:15). Dixon says the approach supports rapid what-if restructuring without repeated Monte Carlo simulation and that source code was provided (30:39–32:25). He describes expected-shortfall extension as work in progress (32:36–34:06).

This supports a bounded partner inference: a public technical interview links an academic/researcher’s risk method to a hedge-fund-facing cloud platform. It does not identify clients or prove client deployment, GenAI/LLM use, model providers, data rights, production permissions, autonomous investment authority, or performance. See the timestamped capture note.

August 28 Risk.net/DZ Bank algorithmic-differentiation and legacy-code route

The recovered Christian Fries Quantcast episode and SoundCloud recording add an adjacent bank/model-risk route. Risk.net and the LMU profile identify Fries as a DZ Bank model-development leader and LMU professor; public finmath documentation provides an implementation lineage.

Fries describes AAD for Bermudan-option and XVA sensitivities and relates backward differentiation to neural-network backpropagation (04:22–09:18). He focuses on runtime scale, parallelism, clean code, understandable logic, and legacy-code integration (14:43–17:00), and says DZ Bank was reimplementing the university code for industry use (23:05–23:38). Follow-up work covered forward sensitivities and approximations for margin valuation adjustment (24:24–25:52).

This is adjacent bank implementation evidence, not proof of any tracked hedge fund’s use or of GenAI/LLM deployment. The sources do not establish model providers, training data, permissions, autonomous investment authority, or performance. See the timestamped capture note.

August 28 Risk.net title-blind Imperial/CFM machine-learning and quant-risk route

The recovered Risk.net Quantcast episode with Damiano Brigo adds a dated academic/industry route that the title filter would miss. The SoundCloud recording was archived and transcribed locally. Risk.net identifies Brigo as Chair of Mathematical Finance at Imperial College London, and Imperial’s faculty profile independently records that position and his CFM-Imperial Institute of Quantitative Finance affiliation.

Brigo describes the research problem as more than model selection. He calls for analysis of black-box ML behavior, convergence, and interpretable outputs (01:40–02:53); points to information-theoretic metrics for finding useful structure in investment firms’ large customer, market, and information datasets (03:09–04:24); and argues for proactive optimization of risk and capital rather than risk measurement alone (04:24–05:48). He also connects quantitative work to regulation, treasury, collateral, systemic risk, and unresolved model-calibration problems (14:29–22:57).

The route is useful for mapping research themes and academic feeders. It is not evidence of a tracked hedge fund’s current AI/GenAI system, model provider, data rights, production permission, autonomous investment authority, or performance. See the capture note.

August 28 Artur Sepp/LGT title-blind quant and applied-GenAI route

The recovered Risk.net Quantcast episode with Artur Sepp adds a named quant and prior systematic-hedge-fund route whose episode title does not mention AI. The SoundCloud recording was archived and transcribed locally. Risk.net’s profile records Sepp’s LGT role and prior systematic-hedge-fund/family-office experience; his current first-party site describes a 10+ person LGT quant team spanning portfolio construction, factor analytics, systematic macro, and applied GenAI.

LGT’s public GenAI/ML Engineer listing describes a portfolio-analysis system and requirements including ML, LLMs, RAG, Python, and financial-data analysis. Sepp’s public repost describes the role as joining his quant team. This is public hiring-intent evidence, not proof of a filled role, model provider, data rights, evaluation, production deployment, or investment authority.

His public GitHub profile exposes ten Python packages spanning analytics, factor modelling, portfolio construction, stochastic volatility, trend-following, private-asset analysis, and Bloomberg data access. The RCM Alternatives transcript identifies an earlier Quantica Capital systematic trend-following route. The Risk.net audio itself covers volatility modelling, risk decomposition, multi-asset portfolio construction, and DeFi interaction risk (00:45–06:27; 15:37–17:08; 38:27–45:09).

This route connects current applied-GenAI hiring language, an open quant-code surface, and dated systematic-investing experience. It does not establish that LGT uses Sepp’s public packages in production or that any historical method transferred to LGT. See the capture note.

The unfiltered Risk.net pass also recovered Patrick Hagan’s Quantcast episode and SoundCloud recording. Risk.net’s dated materials identify Hagan with Gorilla Science and a prior J.P. Morgan CIO quant role. He discusses convexity adjustments, volatility models, calibration, scenario risk, and hedge-fund counterparty opacity (01:27–18:34; 18:34–29:57). The episode contains no AI or GenAI disclosure and is retained as a negative control: technical finance media can expose model-risk and hedge-fund context without providing evidence of AI use. See the capture note.

August 28 Risk.net/Danske Bank differential-ML route

The recovered Risk.net Quantcast episode with Antoine Savine and Brian Huge adds a dated bank research and implementation route whose title does not expose machine learning. The SoundCloud audio was recovered and transcribed locally. Risk.net identifies Huge with Saxo Bank and Savine with Superfly Analytics at Danske Bank; the related differential-ML article and public GitHub organization expose the paper and reproducibility surfaces.

The speakers describe pairing AAD-generated pathwise sensitivities with neural networks and differential PCA to approximate derivative pricing and risk analytics. Named applications include XVA, CCR, SIMMVA, backtesting, risk heat maps, dynamic reports, and trader-workstation computation (02:35–15:12). Savine says differential ML was being deployed for XVA at Danske Bank during the September 2021 recording (16:33–18:34). This is dated, speaker-reported bank workflow evidence for risk analytics, not evidence of a hedge fund’s AI system, alpha, current deployment, provider, proprietary data rights, or performance. See the capture note.

August 28 Risk.net/Pine Tree Market Neutral personnel route

The recovered Risk.net Quantcast episode with Jan Rosenzweig adds a title-blind tracked-manager route. Risk.net identifies Rosenzweig as a portfolio manager at Pine Tree Market Neutral and a visiting senior research fellow at King’s College London; in the recording he describes Pine Tree as a small equity market-neutral quantitative fund and says he is one of two portfolio managers. He discusses fat-tailed and extremal portfolio optimization, independent components, focused hedging, and the distinction between an LDI funding gap and a portfolio-optimization failure (01:26–20:13). The episode contains no AI or GenAI disclosure and is retained as personnel and methodology evidence, not as evidence of Pine Tree’s AI systems, production status, data, permissions, or performance. See the capture note.

August 28 Risk.net/Fidelity title-blind inverse-RL route

The recovered Risk.net Quantcast episode with Igor Halperin adds a dated Fidelity-linked research route whose title does not expose its machine-learning content. The SoundCloud audio was recovered and transcribed locally. Risk.net’s author page and episode page identify Halperin with Fidelity quantitative research and its AI Center for Asset Management at the time of recording. The episode is distinct from the already captured Curious Quant interview.

Halperin describes inverse reinforcement learning as inferring a reward function from portfolio positions and trades, followed by reinforcement learning that seeks to improve on the observed manager behavior (01:33–05:43). The example works at the sector-allocation level and applies a separate momentum rule for stock selection (06:32–08:10). The claimed improvement is a backtest over a relatively homogeneous manager group, not a live Fidelity result (09:21–10:14).

He says the pipeline avoids neural networks to preserve interpretability and describes it as continuing research that might later assist fund managers (13:04–15:40). He later names multi-agent RL and tensor networks as future research directions (32:52–38:30). This is methodology and personnel evidence, but it does not establish a current Fidelity system, production deployment, model/provider inventory, data rights, autonomous permissions, or performance. See the capture note.

August 28 QuantZ/QMIT: an explicit quantamental and agentic-AI surface

The title-blind recovery around Milind Sharma adds a distinct manager-owned media route. QMIT’s April 14, 2026 post about a CQA Las Vegas recording identifies Sharma as the presenter and says the talk covers QMIT’s ML-enhanced ensembles, Enhanced Smart Betas, and a “Hedge Fund in a Box” market-neutral index before turning to LLMs, agentic goal-seeking systems, alignment, and possible AGI effects. The recording’s public automatic captions were recovered and a second timestamped ASR route was checked on August 30, 2026. The capture note preserves both provenance paths and the audio-verification boundary.

The recovered later section adds a bounded automation distinction: Sharma discusses possible GenAI use in compliance, settlement, research-flow automation, and earnings/Fed-call language analysis, while questioning the data volume available for deep learning at medium and low frequencies and the latency fit of GenAI for high-frequency execution (approximately 28:44–31:08). He then contrasts deterministic scheduled quant jobs with agents that may produce different answers to repeated prompts and raises reproducibility and fiduciary concerns (approximately 32:49–34:27). These are presenter-level views, not evidence of QMIT production controls, model permissions, or an investment result.

The current QMIT home page names Sharma as CEO and lists Pratik Sharda as COO, Esma Gregor as Head of Business Development, Sandy Warrick as CRO Emeritus, and Econophysica Ltd. as an affiliate team. The CFA Institute profile separately identifies Sharma as CIO of QuantZ Capital and CEO of QuantZ Machine Intelligence Technologies, and records Carnegie Mellon computational-finance and applied-mathematics training plus a logic doctoral-program connection. This resolves a public personnel and lineage route, not a current headcount or internal reporting chart. The root QMIT.ai page shows a different roster, listing Amit Sardar as CTO and Dr. Oleg Kolesnikov as Head of Research rather than the /home-1 page’s COO and business-development entries. Both pages carry a 2024 copyright notice without role dates, so the discrepancy is recorded as unresolved temporal evidence. The Hedge Fund in a Box page displays June 28, 2024 five-year daily-rebalanced and 22+ year summary charts labeled “19Y backtest with 5 YR LIVE”; those labels document public product positioning, not independently audited performance or model disclosure. The Wiley publisher page confirms Sharma’s March 2026 book on factor investing in the age of machine learning. Its public Google Books contents listing names chapter themes spanning HFIB, LBO and NLP sentiment models, causal methods, agentic AI, agent-driven quant-fund design, portfolio monitoring, and agentic factor investing. These are disclosed research themes and intended chapter coverage; they do not establish that the methods are live, autonomous, or independently tested.

The QMIT media archive exposes a backlog of related source surfaces: the 2022 Dimitri Bianco interview, a 2022 Re*Work AI presentation on “Hedge Fund in a Box,” an IBKR factor-investing master class, a 2020 AI-and-data-science-in-trading panel, a 2019 ML/smart-beta/LBO-prediction presentation, and a 2019 RavenPack presentation on quantamental signals with sentiment. A QuantStrats USA 2025 agenda also lists Sharma moderating a fixed-income/FX panel on AI and LLM risk. These are useful discovery and public-vocabulary signals. They do not disclose model versions, training data, provider relationships, evaluation fixtures, permissions, production status, autonomous capital authority, or performance. The capture and failure details are in the source note. The EQDerivatives Global EQD 2026 programme provides a second conference surface: it lists Sharma for a May 20–21 academic keynote and names public personnel routes at Balyasny, Man AHL, Paloma Partners, a Millennium platform company, and Fidelity in the same event network. Its separate upcoming-event list names Asia EQD on October 27–28, Investor EQD Abu Dhabi on November 18–19, and Japan EQD on November 30, 2026. This broadens the conference watchlist; it does not establish that any listed firm uses QMIT methods or disclose a recording, model, data source, permission map, or result. The Asia EQD 2026 agenda makes the regional route more specific. The October 27–28 Hong Kong event lists an AI-and-investment-decisions session with April Fu of Ping An of China Asset Management Hong Kong and Daniel Xystus of AOP Capital, alongside personnel from Jain Global, Capula Investment Management, Convex Strategies, Life Asset Management, and China Re Asset Management. This is an upcoming programme and personnel surface; no attendance, recording, shared method, model, data source, or outcome is inferred. See the capture note.

August 28 Analyzing Alpha: Tom Starke’s public ML and execution methodology route

The title-blind Analyzing Alpha interview with Dr. Tom Starke, updated March 8, 2024, adds a practitioner and researcher route that is not attributable to a named hedge fund. The publisher page identifies the episode’s AI-trading, portfolio-management, and paper-discussion segments. QuantInsti’s faculty profile identifies Starke as CEO of AAAQuants, records his Nottingham physics PhD and engineering/fluids training, and lists prior proprietary-trading and engineering experience. The YouTube recording was recovered with timestamped captions.

Starke separates predictive modelling, portfolio construction, and execution (17:51–22:28), and describes execution quality as a distinct implementation concern. He frames AI as a tool that should support an investment idea, discusses nonlinear relationships in price data, and distinguishes standard reinforcement learning from deeper models in latency-sensitive execution (23:15–27:58). He says he builds backtests in Python and keeps signals, portfolio sizing, and execution separate (28:02–30:27). He also describes AAAQuants consultancy work and quantitative-alpha modelling for an unnamed large financial institution (41:04–43:59). This is public methodology and lineage evidence; it does not establish a tracked firm’s model, current employer, training data, provider, permissions, deployment, or performance. See the capture note.

August 28 Nassau Street Partners: adjacent capital-advisory workflow signal

The title-blind Nassau Street Partners interview with Gary Shields, uploaded June 29, 2026, adds a current finance-operator route outside the hedge-fund universe. The firm’s LinkedIn post and first-party website identify Shields as Chairman and Managing Partner and describe Nassau Street Partners as a capital-advisory business.

In the video, Shields says the firm uses AI extensively and that some repetitive work can be completed in minutes rather than days (02:35–04:18). He connects the use case to research and capital-advisory work, while separately discussing AI infrastructure and M&A themes (04:05–05:52). The public record does not identify the models, providers, datasets, benchmark, staff denominator, or governance controls. This is a first-party adjacent workflow signal, not evidence of a hedge fund’s trading system, model permissions, autonomous investment authority, or performance. See the capture note.

August 28 Australia: Roger McIntosh, IFM quantitative equities, and academic collaboration

The title-blind ExpertGate interview with Roger McIntosh, uploaded May 15, 2025, adds an Australian institutional-quantitative route. AICRO’s public profile identifies McIntosh as Investment Director at IFM Investors and records La Trobe mathematical-statistics/applied-mathematics training and a University of Melbourne quantitative-finance master’s degree. Financial Standard independently describes his December 2024 appointment to IFM’s quantitative-equities team and its portfolio-construction and factor-insight remit.

McIntosh describes factor-based and sustainable portfolio construction, academic engagement, data-provider workflows, and an AI tool being explored by an ExpertGate colleague for web-scraped ESG information (00:59–05:50; 07:04–10:37). The publisher description says he is interested in applying ML and AI to factor ranking for dynamic asset allocation. That is a public research-interest statement, not evidence of IFM adoption, a named model, an agent, a provider, production permission, or performance. The capture note preserves the source and attribution boundaries.

August 28 CFA Society India: quantitative-value practitioner route

The title-blind queue recovered a CFA Society India webinar with Sharad S. Ramnarayanan, uploaded October 17, 2023. The Global Conference of Actuaries biography identifies him as Appointed Actuary at The New India Assurance Company and records earlier investment roles at Aditya Birla Sun Life Asset Management, Pari Washington Company Advisors, and Tactica Capital Management. The Institute of Actuaries of India excerpt independently describes the book and its quantitative-value approach.

The timestamped presentation separates fundamental expectations, valuation, entry price, and portfolio sizing. It discusses Kelly-style sizing with a practitioner-defined risk cap and position pruning as expected return changes (04:25–22:00; 54:00–64:17). This route adds Indian investment-practitioner and former-asset-management lineage, but it contains no AI, GenAI, LLM, agent, or model-provider disclosure. It is not evidence of a current hedge-fund system, production permission, data rights, or performance. See the capture note.

August 28 Capital Horizons: Re7 Capital’s title-blind DeFi route

The queue recovered Capital Horizons’ interview with Evgeny Gokhberg, published December 12, 2025. The video identifies Gokhberg as Re7 Capital’s founder; Companies House records him as an active director of RE7 CAPITAL LTD, while Re7’s February 2026 media post displays Founder and Managing Partner.

The discussion describes market-neutral stablecoin liquidity provision, overcollateralized lending, smart-contract/code risk, and continuous DeFi monitoring (01:27–16:05). Gokhberg says Re7 built a Bloomberg-like interface for DeFi yields, risks, and alerts. The interview later covers tokenisation and an unnamed large traditional financial institution (28:47–40:47). The European Blockchain Convention profile separately describes Re7 Labs’ on-chain risk curation, vault management, and DeFi ecosystem-design remit.

Re7 Labs’ current first-party page adds a named DeFi Ratings framework that evaluates smart-contract and other DeFi risks and feeds ratings into DeFi asset-allocation models. Other Re7 pages describe proprietary data infrastructure, a “Re7 Risk Engine” underpinning an ETH Yield Strategy on Optimism, Re7 Labs’ claimed risk-curator role in a BUIDL DeFi integration, a Dune co-authored tokenized-funds note, and a Zodia Custody partnership for on-chain fund-interest representation and off-exchange settlement. These disclosures are useful for mapping risk tooling, data, and rails; they do not establish AI/ML, model providers, or autonomous investment authority.

This is a digital-asset strategy and operating-model route, not an AI finding. The reviewed sources do not establish AI, GenAI, LLMs, agents, model providers, training data, production permissions, or independently audited performance. The unnamed institutional relationship is not resolved. See the capture note.

August 28 Hedgeye: historical Millennium/Critical Mass personnel route

The queue recovered Hedgeye’s July 29, 2020 Real Conversations episode with Michael Taylor. The publisher identifies Taylor as a Critical Mass portfolio manager, and the recording’s introduction places him in a Millennium healthcare-manager context (00:10–00:31). The discussion covers healthcare investing, intermediate-term catalysts, mentoring, idea exchange, and risk-managed exits (01:41–05:20; 10:49–12:18; 43:01–45:57). A later Simplify prospectus separately records his Simplify and Critical Mass roles.

This is dated personnel and process evidence, retained as a title-blind negative control. It contains no AI, GenAI, LLM, agent, model-provider, training-data, automation, current Millennium, or performance disclosure. See the capture note.

August 28 Cerebellum Capital: historical automated-ML fund route

The queue’s Cerebellum Capital video lead was rechecked and its 83-second public automatic English caption layer recovered. The clip introduces David Andre with Cerebellum Capital and describes a high-frequency/quantitative conference, increased collaboration, low-latency algorithms, and growing use of statistics and big data (00:00–01:20). It is historical conference context, not a full interview or a current fund disclosure. The David Andre biography and public fund-brochure record nevertheless describe an early finance-native ML process: automated strategy discovery, testing for generalization to unseen data, strategy combination, model-led trade decisions in most cases, and a human-judgment middle ground.

The public Steven Pav CV adds backtest, execution, and research infrastructure; multi-factor models, genetic programming, clustering, regression; execution-impact calibration; and overfit-bias correction. Andre’s Hertz profile supplies Stanford/UC Berkeley and reinforcement-learning lineage and dates his CEO/CTO tenure through 2020. This is a historical AI/ML-fund and personnel record, not evidence of current deployment, GenAI/LLM use, model providers, data rights, or audited performance. See the capture note.

August 28 Optiver: title-blind low-latency infrastructure route

The title-blind queue recovered the Optiver-linked CppCon talk by David Gross, while direct YouTube recovery returned LOGIN_REQUIRED. The official CppCon page and Optiver article identify Gross as an Options Tech Lead and describe a low-latency C++ trading-systems talk. A public transcript mirror makes the content searchable but is not an official timecoded transcript.

The talk covers order-book representation, cache locality, vector and binary-search choices, bounded shared-memory queues, process isolation, profiling with market-data distributions, and the interaction between FPGA paths and flexible software. It also treats time to market as a latency dimension. The content is infrastructure evidence, not a model or strategy disclosure. Optiver’s separate AI Lab research page and 2026 agentic-SDLC article are retained as separate dated records; neither should be inferred to describe this 2024 talk. No reviewed source establishes an AI model, LLM provider, training data, agent order authority, permission map, or independently audited performance. See the capture note.

August 28 Callanish Capital: title-blind historical quant-manager route

The queue’s OpalesqueTV interview with Eoin Murray is dated August 2, 2011 and identifies Murray as CEO and Chief Risk Officer of Callanish Capital Partners, a London-based quantitative hedge fund. Direct recovery now yields a chapter-only VTT, not a complete transcript: the markers cover Murray’s prior CIO role, Callanish’s team and IMQubator/APG seeding, and broad strategy/model/risk topics. The Opalesque roundtable, contemporaneous industry report, and HSTalks catalogue provide the usable cross-checks: Callanish’s 2008 formation, a May 2010 systematic global-macro launch seeded by IMQubator/APG, startup infrastructure challenges, and historical factor, volatility, liquidity, and crowded-trade vocabulary. Later Federated Hermes disclosure supplies a separate chronology for Murray’s subsequent roles.

No reviewed source identifies AI, GenAI, LLMs, autonomous agents, model providers, training data, or AI-attributed results. This is historical quant-manager and operating-context evidence, not current Callanish, GSA, or Federated Hermes technology evidence. See the capture note.

August 28 Odd Lots: allocator view of multi-strategy operating models

The queue also recovered Bloomberg’s Odd Lots episode with Albourne partner Ronan Cosgrave, published May 26, 2025. YouTube was blocked by login protection, while Apple Podcasts, Bloomberg, and timestamped transcript indexes at Tapesearch and Podscripts preserve the episode route.

Cosgrave describes diligence across people, trades, business model, risk model, and investment model (04:28–06:22), and distinguishes traditional multi-strategy structures from pass-through/pod-shop structures while discussing compensation, fees, leverage, cost control, and alignment (18:14–20:29; 28:41–30:12). The episode provides allocator context around named managers, not first-party technology evidence about them. It contains no identified AI model, LLM, agent, provider, training corpus, data-rights, or AI-performance disclosure. See the capture note.

Quantmatic Capital: explicit AI-agent claims in an emerging FX manager

The title-blind queue recovered a Meet the Fund Managers interview with João Guilherme Cruz, whom the interviewer identifies as an industrial engineer building Quantmatic Capital in Brazil. The interviewer’s LinkedIn post, Cruz’s public profile, and the Quantmatic company page provide identity cross-checks. The The Org record displays a cofounder/managing-partner title but marks itself unverified.

Quantmatic’s current first-party website says proprietary AI algorithms, quantitative models, machine learning, and AI agents drive FX signal generation, position sizing, and risk management. It describes agents searching web, social, and news data for fundamentals, sentiment, sudden-change indicators, and event volatility. The page says real capital has been traded since 2025, a team monitors operations with escalation and a kill-switch, and engineering reviews algorithms monthly. The public page does not name model families, vendors, training data, data rights, evaluation, or order permissions.

The interview’s timestamped ASR supplies a second, more granular layer: MetaTrader expert advisors, three algorithms across seven strategies and 21 signals (06:48–07:21); nearest-neighbor language, grid positioning, and global-news/sentiment scraping (07:51–08:17); and an AI copilot suggesting operation timing and lot size while a person remains in the operating loop (16:02–17:15). The interview dates coding and structuring to 2024 and live operation to 2025 (04:02–04:36).

The current website says “without human discretion,” whereas the interview describes a human-operated copilot. That is a material unresolved temporal or definitional discrepancy, not evidence that one description is correct. The capture note records both surfaces and does not infer model autonomy, performance, or current team ownership.

August 28 Plettenberg Capital: a software-factory architecture with an explicit AI boundary

The Sophron episode identifies Alexander Levin as Founder and Chief Investment Officer for public markets and Daniel van Flymen as Co-Founder and Chief Technology Officer at Plettenberg Capital. The publisher description names roughly 500 live algorithms, a common research/simulation/live-execution engine, point-in-time and corporate-action checks, in-house execution, layered safety controls, and a stated absence of language models near market decisions. The recovered audio is being treated as a source for spot checking, while the initial article treatment remains description-level evidence.

The stated AI role is workflow compression: the publisher places it in research discovery, testing, and validation, while describing price-focused research and the deliberate exclusion of volume, fundamentals, and alternative data in the discussed process. This exposes a design choice between AI-enabled research tooling and a non-LLM market-decision path. It does not reveal model families, data vendors, simulator implementation, evaluation fixtures, trade permissions, or production controls.

The recovered audio makes the boundary more specific. At approximately 24:17–30:20, the speakers describe about 500 production algorithms, decision-tree and allocation layers, point-in-time prices, ticker changes, and corporate actions. At 43:08–44:47, they say AI is used to vary hypotheses, triage backtest/reporting anomalies, and accelerate visualization, while not measuring the market, touching an order, or sitting in the production line. They describe deterministic outputs and traceability to the decision-tree path. At 50:04–50:50, they describe highly automated monitoring and execution, with people still monitoring the system and making limited manual changes. These are speaker statements from recovered audio, not independently verified deployment or permission evidence.

The public record also contains a potentially important time/scope distinction. A van Flymen LinkedIn post describes internal tools and an autonomous engineering platform; the accompanying Levin post says an internal Plettenberg LLM supports natural-language interaction with portfolio and returns. That can coexist with the episode’s “no language model near the market” description if the LLM is an internal interface rather than a signal or execution component, but the public evidence does not prove that interpretation. Levin’s launch post names Clear Street, Middlegate, Riveles Wahab LLP, NAV Fund Services, and U.S. Bank as operating partners without public scope or data-access detail. See the capture note.

August 28 Alex Zhong: validation and portfolio construction across quant platforms

The Sophron episode and publisher announcement describe a practitioner route through WorldQuant’s BRAIN/ACE environments, the DRW Crypto Predict competition, and Trexquant machine-learning equity research. The publisher frames the episode around out-of-sample validation, signal combination, alpha decay, market-structure change, and a portfolio of more than 50 crypto strategies. A public profile displays AX Quantitative and current crypto/quant hiring, but the captured profile does not independently verify all historical affiliations.

The signal here is methodological rather than a disclosed production model: a research pipeline must survive out-of-sample checks, account for correlation when combining signals, and reduce exposure as alpha decays. No reviewed source establishes an LLM, agent, current employer mandate, model family, training data, execution authority, or independently audited performance. See the capture note.

August 28 Shell / Nikita Granger: published RL routes for gas and power decisions

The Sophron episode identifies Nikita Granger as a Shell Quantitative Developer and describes work spanning algorithmic trading, energy markets, and structured derivatives. Shell’s official research bibliography lists Granger as a co-author on two 2023 KDD Workshop papers: deep RL for natural-gas futures and domain-adapted/imitation-based DRL for Dutch power arbitrage. A public alumni profile provides the visible econometrics, applied mathematics, and computational-science lineage.

The gas paper frames the task as sequential trading in natural-gas futures and discusses domain adaptation, interpretability, and an ensemble intended to address stability and turnover. The power paper frames day-ahead/intraday arbitrage as a dual-agent RL problem, adds domain-informed imitation and reward engineering, and models order tranching. These are explicit model objectives and published research artifacts. The reported comparisons remain paper-level results; no reviewed source establishes live Shell production, current ownership, data rights, execution permissions, or financial performance after live costs and controls. Shell’s corporate AI page is recorded separately and is not attributed to this named-person route. See the capture note.

August 28 The Sophron Network: autonomy, agentic market intelligence, and historical AHL context

The Sophron Network RSS feed, its Apple listing, and direct Spotify/Anchor episode pages add a publisher route that was not in the earlier queue. The feed exposed direct enclosures, so the three relevant episodes were downloaded and processed with temporary local English MLX ASR on August 28. The timestamped recovery note records bounded paraphrases; full third-party transcripts remain private.

The Jeremy Kadouch episode identifies him as Portfolio Manager at MN Fund. MN Fund’s first-party page separately displays that role and a remit covering strategies and capital deployment. The recovered audio places AI in stack construction and external research enhancement, while Kadouch says the executable strategy reads market data and applies mathematical rules without AI inside the execution path (16:56–23:30). The episode also describes a separate Quorum Index project with an OpenClaw manager, sub-agents, recurring collection, and a possible LLM-powered terminal (52:17–56:34). These are public statements about two distinct scopes, not proof of MN Fund’s internal architecture, evaluation, or live permissioning.

Kadouch’s separate public Quorum Index post describes continuous crypto-pair scans, OHLCV/narrative/macro inputs, OpenClaw routing and memory, 22 scheduled skills via the Anthropic API, weekly Claude Code review, and an MCP server for market state, signal reliability, and output quality. Other public posts describe repetition checks over the last five outputs and event-driven repair with cooldown and scope limits (feedback, repair). The record labels this as a separate public project associated with Kadouch. It does not attribute those components to MN Fund or establish trading authority, model evaluation, or performance.

The Robert Carver episode supplies a historical Man AHL personnel route and a current independent-system route. The publisher describes seven years at AHL, a most-recent Head of Fixed Income role, and a personal automated system covering about 250 futures markets using pysystemtrade. In the recovered audio, Carver discusses AI as a possible shortcut that can also encourage overfitting, prefers robustness across regimes to heavily parameterized regime classification, explains a modular simulation/production/broker architecture (49:51–53:17), and warns against trusting AI- or “vibe”-coded trading software with money (65:19–68:05). No reviewed item shows that AHL used Carver’s current personal stack or that a named AI method was deployed there.

The Amay Patel episode identifies a Partner and Head of Investments at Atomic Digital and describes a 2025 market-neutral digital-asset fund across OTC structured products, DeFi, quantitative, fixed-income, and credit markets. Local ASR closes the audio gap and captures one isolated GPT reference in a legal-knowledge context (13:17), but no Atomic Digital AI system, model, data pipeline, or agent authority. See the full capture note and ASR recovery note.

August 28 Binomial Technologies: public models, open artifacts, and a human approval gate

Binomial’s first-party research page describes fine-tuned open-weight LLMs, classifiers, forecasters, and task-specific networks trained on the firm’s own data pipeline. Its Sentry page describes a supervisor routing work to separately prompted specialists for equity research, quant/risk, derivatives, capital flows, and market signals, with persistent workspaces, tool/citation trails, scheduled tasks, Excel, and SDK surfaces. The Binomial Marks model card adds a public 400M-parameter earnings-call scorer with 23 structured outputs, a 16,384-token context, a claimed 80,000+ transcript training set, and a held-out 2,000-call comparison against frontier-model labels. It labels the model a research preview and not an alpha model.

The firm’s Confidence Premium note describes a peer-relative confidence factor derived from Marks and reports a 63-day rank IC of +0.025 across a July 2011–July 2026 evaluation sample. The note also reports hypothetical costed portfolio results. These are dated, firm-authored backtest claims, not independent replication or live-return evidence. The separate Shannon 2 note describes an approximately 150M-parameter news model that routes ticker and macro articles and emits structured event, direction, novelty, specificity, materiality, and claim-type outputs. This gives the public record two distinct model lanes—earnings-call language and news—while leaving training-data rights, point-in-time joins, production permissions, and live signal contribution unresolved.

The PyPI BinomialHash artifact provides a separate open-software route for structured-data compaction, retrieval, statistical/causal tools, and provider adapters. Binomial’s firm page also claims an OpenAI pilot, but no independent OpenAI confirmation or contract scope was found. Full source boundaries are in the Binomial capture note.

August 28 title-blind routes: Askeladden, enterprise controls, and Boston quant history

The MOI Global interview with Samir Patel identifies him as founder and portfolio manager of Askeladden Capital, a long-only public-markets manager. Patel describes source-grounded orientation, deeper investigation, due diligence, archive retrieval, thesis monitoring, and accounting review, with human source checking and materiality judgment retained (09:36–14:57; 18:11–24:48; 45:56–52:01). He names personal use of general-purpose models, but the episode does not establish an enterprise contract, fine-tuning corpus, model registry, autonomous allocation, or performance. The capture note treats automatic captions as navigation evidence.

The Odd Lots interview with Goldman Sachs CIO Marco Argenti is an enterprise-control comparison point, not a hedge-fund deployment record. Argenti describes a GSAI Assistant used by approximately 47,000 people, hundreds of data sources, a lakehouse/MCP path, agentic developer tooling, token metering and model gateways, provider-side forward-deployed engineers, and production controls (06:25–11:10; 19:52–26:51; 38:28–40:33). The statements are executive-reported bank-technology claims with automatic-caption navigation and should not be transferred to Goldman Asset Management or a hedge fund. See the capture note.

The 2016 Boston Data Mining talk by Sean Kruzel identifies him in the event description as Astrocyte Research’s founder and CEO, with prior quantitative global-macro and fixed-income-relative-value hedge-fund work and MIT mathematics/economics training. Its historical methodology route covers scientific practice, machine learning, trade sizing, fat tails, and strategy evaluation. The captions are noisy and require audio review before quotation; the route does not establish Astrocyte’s current status, model, data, deployment, or performance. See the capture note.

August 28 title-blind routes: Askeladden, enterprise controls, and Boston quant history

The MOI Global interview with Samir Patel identifies him as founder and portfolio manager of Askeladden Capital, a long-only public-markets manager. Patel describes source-grounded orientation, deeper investigation, due diligence, archive retrieval, thesis monitoring, and accounting review, with human source checking and materiality judgment retained (09:36–14:57; 18:11–24:48; 45:56–52:01). He names personal use of general-purpose models, but the episode does not establish an enterprise contract, fine-tuning corpus, model registry, autonomous allocation, or performance. The capture note treats automatic captions as navigation evidence.

The Odd Lots interview with Goldman Sachs CIO Marco Argenti is an enterprise-control comparison point, not a hedge-fund deployment record. Argenti describes a GSAI Assistant used by approximately 47,000 people, hundreds of data sources, a lakehouse/MCP path, agentic developer tooling, token metering and model gateways, provider-side forward-deployed engineers, and production controls (06:25–11:10; 19:52–26:51; 38:28–40:33). The statements are executive-reported bank-technology claims with automatic-caption navigation and should not be transferred to Goldman Asset Management or a hedge fund. See the capture note.

The 2016 Boston Data Mining talk by Sean Kruzel identifies him in the event description as Astrocyte Research’s founder and CEO, with prior quantitative global-macro and fixed-income-relative-value hedge-fund work and MIT mathematics/economics training. Its historical methodology route covers scientific practice, machine learning, trade sizing, fat tails, and strategy evaluation. The captions are noisy and require audio review before quotation; the route does not establish Astrocyte’s current status, model, data, deployment, or performance. See the capture note.

August 28 BedRock Partners: methodology-specific AI and historical buy-side personnel claims

The first-party BedRock E19 transcript identifies Bill Sun Qingyun and records his statements about Stanford mathematics training, early Google Brain work, and historical roles at Millennium, Citadel, and Point72. Those employment details are guest-reported and remain unverified in this source note. The separate Cong Tan founder profile supplies a first-party Stanford/Fudan lineage and a standardized research-framework description.

The discussion is useful because it draws a boundary around methodology-specific assistants: repeatable, back-testable tasks are presented as more amenable to automation; longer-horizon judgments about unprecedented events require ongoing human interpretation. The speakers describe retrieval, key-point extraction, private-data and expert-call access, methodology templates, and Bayesian updating. A separately mentioned Gen Alpha / AIUSD.AI crypto product is retained as a product lead, not as evidence of BedRock deployment or regulatory status. BedRock’s archive also exposes older ML and cybernetics reading routes. See the BedRock capture note.

The cross-surface expansion adds a Beacon Pin AI profile plus Chinese-language Bilibili and Indigo Talk discovery surfaces. These broaden the personnel and regional-media search route, but the repost, podcast metadata, and secondary reporting do not independently establish employer deployment, product safety, regulatory status, or live performance.

The adjacent BedRock archive now has a recoverable public-audio route for E25–E30 through Xiaoyuzhou, which filled the timestamp gap left by unavailable YouTube captions. E28’s automatic Chinese-ASR navigation points to discussion of retrieval, model updates, data tracking, basic research agents, monthly AI-use reflection, and researchers acting as product managers for AI collaboration (approximately 00:00–01:00; 07:36–08:14; 08:36–15:13). E25 discusses analyst-defined questions and human judgment around AI-assisted research (approximately 41:09–47:14); E26–E27 cover external engineering, agent environments, inference, and adoption; E29–E30 are client/manager discussions of AI, infrastructure, and investment frameworks. Earlier E16, E20, E22, E23, and E24 add public discussion of cloud/model economics, AI-investment themes, agents, parallel model updates for earnings work, and young-analyst workflows. These are timestamped paraphrases from automatic ASR, not verbatim quotations. They do not establish a model inventory, data rights, production deployment, agent permissions, autonomous investment authority, or performance. See the recovery note.

August 28 AIMA Singapore: fund-manager briefing and title-blind personnel route

The AIMA Singapore Fund Manager Briefing page records a January 21, 2026 in-person event framed around foundation models, deployment, governance, and measurable outcomes in finance. The displayed speaker list includes Morgan Stanley’s Natalie Ang in Prime Brokerage Business Consulting; Hozefa Topiwalla, Head of Asia Research Product and Asia Lead for Firmwide AI; Matt Zhang, Chief Risk Officer at Keystone Investors; and Gary Ang in AI Governance and Risk Management. This expands the Singapore personnel and event route beyond AI-labelled job titles.

The page provides agenda and displayed-role evidence only. No replay, transcript, model inventory, dataset, vendor, production permission, investment authority, or AI-attributed performance was found. The event should therefore be treated as a discovery and personnel lead, not as evidence of a shared Morgan Stanley/Keystone implementation. See the capture note.

The public LinkedIn profile for Matt Zhang Changhao separately displays a Keystone Investors affiliation and lists him as a panelist at the briefing. It also exposes an NUS education record and a professional certificate in investing using machine learning and alternative data. A public profile for Hozefa Topiwalla associates him with Morgan Stanley in Asia, but does not reproduce the AIMA title. These are personnel and discovery corroborators, not evidence of a model, dataset, production authority, or investment outcome.

August 28 AIMA: AI meeting tools as a current title-blind route

The AIMA event page records an August 26, 2026 webinar with Charu Chandrasekhar of Debevoise & Plimpton and Suzan Rose of AIMA. The organizer frames the session around AI-generated transcripts and summaries of calls and meetings, changing adoption, compliance questions, and fund-manager considerations. This is a useful discovery route for voice, meeting-data, retention, and governance research even though the event title does not name a hedge fund or an AI model.

The public page exposes event metadata and a replay link, but the replay article was not publicly retrievable in this capture. No transcript, fund-specific implementation, vendor or model inventory, recording policy, production permission, investment authority, or performance claim is promoted. See the capture note.

August 28 Abu Dhabi Finance Week 2026: Middle East conference and speaker route

The ADGM announcement and ADFW official site schedule Abu Dhabi Finance Week 2026 for December 7–10, 2026. The announced first wave includes Dmitry Balyasny, identified as Co-Founder, Managing Partner and Chief Investment Officer of Balyasny Asset Management, and Steven Desmyter, identified as President of Man Group. ADGM also announces an inaugural AIMA Global Hedge Fund Leaders Platform. This adds an English/Arabic regional conference route and a concrete follow-up surface for public recordings, agendas, and speaker posts.

The source establishes event and displayed-role metadata only. It does not establish attendance, what either speaker will say, AI or GenAI implementation, model/provider choice, data partnerships, permissions, deployment, investment authority, or performance. The source note preserves that boundary.

The MSCI Institutional Investor Forum Doha page schedules a September 14, 2026 MENA event whose agenda includes AI, geopolitical change, portfolio construction, public-market factors, and private-market data. It lists Ashley Lester, Seibert Kruger, Uday Karri, and Bentley Kaplan in MSCI research or institutional-investment roles. This is a vendor conference-monitoring route, not a hedge-fund deployment disclosure; no transcript, attendee confirmation, model, dataset, permission, or performance evidence is available. See the source note.

August 28 Arthur AI Fest: model-evaluation and Balyasny speaker route

The Arthur AI Fest page lists Charlie Flanagan as Head of AI at Balyasny Asset Management alongside an agenda covering LLM evaluation, benchmark design, inference efficiency, control vectors, embeddings, AI platforms, agents, multimodal models, and governance. The page displays September 26 but does not state the event year. It is therefore a vendor-conference speaker and topic route only; it does not establish what Flanagan said, a Balyasny/Arthur partnership, a model or dataset, deployment, permissions, or performance. See the source note.

August 28 UBP Hedge Fund Conference: AI panel and manager-discovery route

The UBP conference recap, dated July 17, 2026, says its London Hedge Fund Conference had more than 170 attendees and included clients and managers from the UK, Asia, Europe, and the Middle East. UBP says representatives of Campbell, Brigade, and Shannon River participated in panels that discussed artificial intelligence alongside macro and liquid-credit themes. UBP names Kier Boley and John Argi as Co-Heads of its Alternative Investment Solutions platform. The embedded video is cookie-blocked and no individual manager speakers or transcript are exposed. This is a first-party event metadata route only; it does not establish any manager’s model, dataset, partnership, deployment, permissions, authority, or performance. See the source note.

August 28 Institutional Investor FTF: CIO/CTO and Automating Alpha route

The Institutional Investor FTF Winter Workshop & Automating Alpha page displays February 24–25, 2026 in Fort Lauderdale and describes a buy-side CIO/CTO forum focused on artificial intelligence, new data sources, and analytical tools in investment management. The page says registration is not open and provides no complete public agenda, buy-side speaker list, transcript, or recording. It is therefore an event-discovery route for title-blind technology and investment-workflow evidence only, not evidence of attendance, a named fund’s model, data partnership, deployment, authority, or performance. See the source note.

August 28 J.P. Morgan QIS: JPMaQS, hedge-fund overlays, and bounded LLM workflow use

The first-party J.P. Morgan Making Sense episode on QIS, published November 11, 2025 and recorded October 15, 2025, names Arnaud Jobert as co-head of Strategic Indices for Markets and global head of Equity Structuring and Eloise Goulder as head of the Data Assets & Alpha Group. Jobert describes a QIS platform crossing $100 billion of client notionals, hedge-fund use of execution outsourcing and central-risk-book overlays, and a Nexus platform where clients retain their own models and intellectual property while outsourcing execution. The transcript also describes JPMaQS as a point-in-time macro-fundamental dataset and frames LLM use around factor/performance commentary and reducing time to market. These are provider statements, not named-client or trading-model evidence; no provider, fine-tuning, training corpus, client implementation, permission, or performance result is disclosed. Notionals are not AUM. See the source note.

The related first-party Deepak Maharaj QIS episode, published November 12, 2024 and recorded October 7, 2024, describes an LLM-enhanced Quest thematic-index workflow. The episode says news articles are used to identify company-theme associations and that an LLM refines the search vocabulary. Maharaj describes internal testing against an earlier NLP approach without publishing its design or result, and flags output randomness as a challenge for systematic repeatability. He also describes hedge-fund factor/thematic use cases and a Macrosynergies partnership for macro time-series data, alongside exploratory social-media, sentiment, intraday, and credit-card data work. This remains provider-reported product-development evidence; it does not establish a named client, model provider, data rights, production permission, autonomous authority, or performance. See the source note.

The same first-party archive’s July 2025 intraday-QIS episode identifies Mathieu Boisot as head of Cross-Asset, Volatility and Intraday Product Development. The transcript describes live gamma maps, zero-day-to-expiry options, levered ETFs, intraday data, and daily estimates of options gamma imbalance as inputs to understanding possible intraday momentum or mean reversion. Boisot gives a speaker-reported observation that roughly 60% of relevant volume was less than one week to expiry, without publishing a methodology or denominator. This is a provider market-structure/data disclosure, not evidence of a named client’s model, permission, autonomous authority, or performance. See the source note.

The related first-party Rui Fernandes structuring episode, published February 21, 2025 and recorded January 29, 2025, describes an approximately 140-person cross-asset structuring team, platform interoperability, customization at scale, multi-strategy client needs, and a Nexus structure in which clients retain their models and intellectual property while outsourcing execution. AI is discussed only as a broad emerging trend. The episode provides no model, agent, data vendor, evaluation, named hedge-fund implementation, permission, or performance evidence. See the source note.

August 28 Quant Enthusiasts: Fortitudo’s generative-market-model and open-source route

The title-blind queue also recovered the Quant Enthusiasts interview with Anton Vorobets, founder of Fortitudo Technologies. Fortitudo’s first-party company page describes institutional investment and risk-management software, collaboration with institutional investors, an Investment Analysis module, an Investment Simulation module, and a Data Science Server. Its public company materials link to open-source Python and identify quantitative finance, machine learning, cloud computing, generative AI, portfolio construction, and risk management as relevant specialties.

The interview is unusually explicit about a modality that is separate from LLM text systems. Vorobets discusses applying variational autoencoders and GANs to tabular investment time series and proposes “large market models” for high-dimensional market simulation (17:07–19:27). He describes transfer learning across U.S. and European implied-volatility and interest-rate markets, including surface and curve structure and possible no-arbitrage constraints (18:33–19:27). He also describes selecting methods according to the complexity of the investment problem and prioritizing empirical, computational, and practical data work (19:42–20:54; 21:21–23:29). These are dated practitioner and company statements; they do not establish a named hedge-fund customer, model weights, training corpus, production deployment, permissions, autonomous trading authority, or independently verified performance. See the timestamped capture note.

The current Fortitudo homepage adds a separate product-positioning signal: it places generative AI inside the Investment Simulation module for future return and risk-factor paths, lists asset managers with hedge-fund strategies among the relevant client categories, and describes a Python/AI-ML Data Science Server. The same page explicitly says the public Python package is separate from the proprietary modules and is not representative of their quality or functionality. Related public repositories cover the book’s Python code and CVaR optimization benchmarks, while a Substack VAE post documents a time-series VAE case study. This combination gives the research program an inspectable methodology surface plus an explicit generative-AI product claim, but not a named customer implementation, proprietary model inventory, customer data, production permission, or performance audit.

August 28 registry reconciliation: Bridgeway research leadership and a dated ML boundary

The Excess Returns interview with Andrew (Andy) Berkin was already captured locally on August 18, but it had not been reconciled into the central media registry. The publisher page identifies Berkin as Bridgeway’s Head of Research and links the Apple, Spotify, and YouTube versions. Bridgeway’s August 22, 2024 prospectus supplement independently identifies Andrew L. Berkin, PhD as Head of Research and Portfolio Manager, says he oversees development and implementation of statistically driven, evidence-based strategies, and records a BS in Physics from Caltech and a PhD in Physics from the University of Texas. Those are role and lineage anchors; the prospectus does not describe AI or GenAI.

The recording’s AI/ML segment is near the end of the approximately 59-minute episode. The host introduces examples involving AI-based fundamental forecasting and NLP classification of firms by price versus intangible or brand value (50:09–50:49). Berkin says the mathematics and statistics behind such methods need to be understood while acknowledging interesting work in machine learning, NLP, and AI (50:59–51:49). He warns that additional degrees of freedom can increase data-mining risk (51:51–52:20), then points to asking why a result works, holdout samples, and robustness checks (52:35–53:04). He describes Bridgeway’s work in the area as introductory and says the firm was not heavily invested in it “right now” at the time of the recording (53:04–53:18).

This provides a dated public statement about Bridgeway’s research posture and validation vocabulary, not a current AI strategy or deployment record. It does not disclose a model architecture, provider, training corpus, data rights, agent permissions, production endpoint, investment authority, or AI-attributed performance. The phrase “introductory” is tied to the 2023 recording and should not be projected onto the firm’s current state. The reconciled source note records the corrected timestamps and evidence boundaries.

August 28 title-blind institutional-manager route: Guggenheim’s firmwide AI statement

The first-party Guggenheim Macro Markets Episode 78 with CIO Anne Walsh was found through a generic 2026 investment-outlook title rather than a hedge-fund or AI query. Guggenheim’s page dates the episode December 19, 2025, identifies Walsh as CIO of Guggenheim Partners Investment Management, and publishes a full transcript. The page also links the YouTube, Apple, and Spotify versions.

Near the end of the transcript, Walsh says Guggenheim is deploying AI throughout the firm and integrating it into virtually every segment of the business. She places the stated use cases inside investment processes, risk management, investment reasoning, portfolio construction, and security analysis (27:48–28:25). She then extends the list to trading, compliance, legal, and operations, with stated goals of reducing costs and accelerating timelines (28:25–28:37). She describes team members creating technology solutions and deploying use cases through 2026 (28:37–28:58).

This is useful adjacent evidence of an executive-described operating strategy that spans both investment and control functions. It does not identify a model, provider, training data, budget, named implementation owner, evaluation protocol, production control, agent permission, autonomous investment authority, or AI-attributed performance. It is not evidence of a hedge fund’s system. The source note preserves that boundary.

August 28 title-blind context route: BlackRock’s AI-and-markets outlook

The first-party BlackRock Investment Institute episode “2026 Outlook: Pushing Limits” was found through a generic market-outlook title. The page’s embedded transcript and the Apple listing identify Jean Boivin, Head of the BlackRock Investment Institute, in conversation with The Bid host Oscar Pulido. The page also links the Spotify and YouTube versions.

The publisher’s chapter map places AI’s economic impact at 03:33–05:39, financing the AI build-out at 09:44–12:30, and portfolio-construction implications at 12:30–15:06. The episode frames AI capital expenditure and digital infrastructure as macro variables, discusses leverage needed to finance large projects, and describes the risk that portfolios that appear diversified may contain concentrated exposure to a small set of structural forces. These are BlackRock Investment Institute outlook statements, not a description of BlackRock’s internal AI systems.

This route adds a first-party market-context layer to the discovery graph and confirms that generic outlook titles can carry substantive AI material. It does not disclose a model, provider, training data, research permissions, autonomous investment authority, or AI-attributed performance. The capture note keeps the outlook evidence separate from firm-internal deployment claims.

August 28 publisher-archive expansion: BlackRock Systematic’s fixed-income AI route

The BlackRock AI podcast archive exposes a distinct episode, AI and Bond Markets: How Artificial Intelligence Is Reshaping Fixed Income Investing. The first-party page and Apple listing identify Jeff Rosenberg as Senior Fixed Income Portfolio Manager at BlackRock Systematic and date episode 256 to April 2, 2026. The publisher transcript makes this a separate investment-side route from Jean Boivin’s macro-outlook episode.

Rosenberg describes machine learning and generative AI as part of systematic investing and discusses sentiment analysis across central banks and thousands of issuers (08:42–10:14). He then describes combining systematic breadth with deeper fundamental analysis of earnings releases, press conferences, analyst reports, and media coverage (12:33–14:43). The closing section connects those capabilities to real-time or near-real-time portfolio optimization, liquidity awareness, electronic credit trading, and transaction-cost control (14:43–17:04).

This is named practitioner evidence about a systematic fixed-income workflow, with AI and generative-AI language attached to research and portfolio construction. It does not disclose which tools are ML versus generative AI, a model architecture, provider, training corpus, data contract, evaluation protocol, production status, permissions, autonomous authority, or AI-attributed performance. The source note records the publisher hierarchy and boundaries.

August 28 vendor-media expansion: Weights & Biases quant and agent webcasts

The public Weights & Biases BrightTALK channel surfaced four quant- and agent-relevant webcasts that were absent from the registry. The large-scale agentic quant-research demo describes multi-agent alpha research, tool and model-output visibility, configuration experiments, and evaluation. The new quant-stack session names UCSD researchers alongside a Weights & Biases AI engineer and describes large-scale experimentation, simulation, compute, data, and deployment. The quantitative-trading model-performance session adds a vendor/customer discussion of the research-to-production toolchain, and the agent-swarms session describes a propose–run–evaluate–select loop with tracing, online evaluation, lineage, and quantitative trading as one use case.

These pages add a vendor vocabulary and a capture queue, not evidence that any named hedge fund uses the systems. They do not identify a customer’s model, training data, data rights, production permissions, autonomous authority, or performance. The capture note records the metadata and the recovery boundary.

August 29 title-blind platform route: Foresee Markets

The Foresee Markets podcast was found through generic investing and AI/quantitative search rather than a hedge-fund query. Its public Riverside RSS feed contained 23 episodes on August 29, 2026, each with an audio enclosure and a publisher-provided transcript URL. Selected episodes describe a six-dimensional stock-selection framework, sector-relative ranking, automatic exits, short-horizon market relationships, options overlays, and a planned interface for themed long/short portfolios and later liquidity-provider connectivity. Because the publisher transcripts are plain text, targeted local ASR now supplies navigation for Episode 4’s AI/ML selection and universe filtering (00:46–01:31; 04:08–05:50; 06:12–07:01) and Episode 17’s platform workflow (02:31–05:21) and planned liquidity-provider connection (07:53–08:00). The related capture note records the audio and VTT hashes and keeps the ASR as automatic navigation evidence.

This is useful title-blind evidence about the vocabulary of an algorithmic-investing product, not evidence of a hedge-fund deployment. The show’s backtest, return, accuracy, and automation figures are speaker-reported. The reviewed feed does not provide code, a complete point-in-time universe, transaction-cost assumptions, broker records, model weights, customer identity, or independent performance verification. See the Foresee capture note.

September 1 title-blind podcast routes: Point72, Coatue, and Millennium

The Point72 announcement dated June 3, 2025 links the Becoming a Hedge Fund Analyst: Inside Point72 Academy series to Spotify, YouTube, and Apple Podcasts. It describes a STEM-to-investing episode featuring three analysts and Ariel Herman, identified as Long/Short Analyst Engagement Director with a PhD in biochemistry and molecular biophysics. The described transfer is methodological—hypothesis formation, experiment design, complex-data analysis, and drawing conclusions—not a disclosure of a model or AI system.

The Apple catalog lists 11 episodes from 2022–2025 and identifies Jaimi Goodfriend as Point72 Academy director. Its episode titles create a useful title-blind personnel and operating-model queue: analyst hiring, internship design, international Academy offices, analyst/PM team formation, curriculum design, and interviews with Steve Cohen. The catalog explicitly says the podcast may not be copied or distributed without Point72’s prior written consent. Accordingly, the project records the catalog and distribution routes as metadata and provenance; it does not republish the catalog’s transcript or audio. See the capture note.

The Omny.fm page for Matthew Granade adds an alternate publisher route to the existing historical interview. Its description identifies Granade as Point72’s chief market intelligence officer, says he directed the central portfolio and managed Point72 Ventures, and describes the venture arm’s stated areas as financial technology, enterprise automation, artificial intelligence, and cybersecurity. It also records his Domino Data Lab and Bridgewater lineage. This corroborates historical role and workflow context; it does not establish current Point72 reporting lines, model inventory, permissions, or deployment.

The Michael Recce episode is a separate historical route, published May 24, 2021. The publisher describes Recce’s path from AI academia and advertising technology into the hedge-fund space with Point72, and says he created data-science teams for a hedge fund, Singapore’s sovereign wealth fund, and Neuberger Berman. Its public Acast enclosure was recovered for private timestamped ASR. These are publisher and historical practitioner statements, not current Point72 or Neuberger Berman system evidence.

The Podwise index adds a chapter and timestamp surface to the already captured 2019 Coatue/Point72 data-science panel. Its index describes public/private investing scope, real-time economic data, data sourcing, model-informed research, data-literacy requirements, cloud-first infrastructure, and data-quality/permissioning concerns. It is treated as a derivative navigation surface; the existing YouTube recording remains the canonical media record.

Millennium’s official mChats with Ross Garon page, dated September 2025, identifies Garon as Global Head of Quantitative Strategies. He distinguishes quantitative researchers, who tend to be statistics-trained, from quantitative developers, who tend to build high-throughput research and trading infrastructure. The page also describes firm-provided data, risk models, execution algorithms, technology infrastructure, hiring support, and advice. This is first-party organizational evidence and a useful non-AI title route. It does not name a GenAI model, training corpus, agent permission, desk implementation, or investment result.

The recovered Michael Recce episode adds a historical cost-and-organization lane. Recce says the Point72 plan at the time was to build a 20–30-person data-science team quickly and integrate datasets with investment teams (00:13:27–00:13:39). He separates short-horizon machine-readable applications from quarterly and longer-horizon uses (00:18:25–00:19:15) and says the data-science role begins with understanding the business question before selecting the solution (00:25:22–00:25:35). In a later discussion of Neuberger Berman, he gives a historical estimate of roughly $5–6 million for a data-science team (00:35:12–00:35:20). The recording was recovered as 687 timestamped automatic-ASR segments and archived privately; speaker attribution and wording remain subject to audio spot-checking. These are dated practitioner statements, not current Point72 or Neuberger Berman budgets, model inventories, permissions, deployment evidence, or performance. See the capture note.

September 1 title-blind Algert and Neuberger Berman process disclosures

The North Square April 2026 Algert Global episode page links a publisher transcript and identifies Ryan LaFond as Algert Global’s Co-Chief Investment Officer. In that transcript, LaFond describes machine learning for building and combining signals, LLMs for extracting text features such as whether management answers the questions asked, and a process that evaluates hundreds of insights across thousands of stocks as information changes. He also describes proprietary risk and transaction-cost models inside optimization. This is unusually specific first-party process language, but it still does not name a model, provider, training corpus, evaluation design, data contract, human approval point, or autonomous authority. The capture note records the transcript hash and private-retention boundary.

The first-party Neuberger Berman quantamental transcript, whose PDF footer is dated 2019, identifies Tim Creedon, Ray Carroll of Breton Hill, and host Anu Rajakumar. It describes a joint fundamental/quantitative process, an overnight universe of roughly 6,000 tickers, a central-research data-science team integrated with analysts, and historical use of credit-card, web-browsing, and job-posting data to test investment theses. The transcript does not establish that those roles, datasets, or practices remain current, and it does not disclose an LLM, agent, model architecture, training data, or production permission map. The capture note records that the publisher returned HTTP 429 to the local archival fetch; no private file checkpoint is claimed.

September 2 title-blind quant-researcher expansion: Quadrature, Capula, and ExodusPoint

The MIT CSAIL event record and CSAIL Alliances page identify a Quadrature technical talk held on October 2, 2025. The pages name Daniel Goldbach and Hanna Yakubovich as Quantitative Developers. The event biographies describe Goldbach’s prior Head of ML Infrastructure remit and current focus on low-latency research and trading systems, and Yakubovich’s work across low-latency infrastructure optimization and portfolio construction with a current focus on alpha-forecasting research. No recording or transcript was located. This is public event-biography evidence, not proof of a current model, deployment, or performance. See the capture note.

The Macro Hive publisher archive and episode listing add a title-blind Capula route: “Chris Crowe On End Of Safe Asset Shortage,” dated July 2, 2020. The publisher describes Crowe as Head of Economic & Flow Research at Capula Investment Management, with earlier Barclays and IMF experience. An Apple catalog entry separately identifies him as Capula’s Head of Economic Research and describes G10-and-China coverage. The route contributes historical personnel and macro-research metadata; it does not add Capula AI or GenAI evidence, and no audio or transcript capture is claimed.

The Risk.net paper identifies Jianfei Zhang as a quantitative researcher at ExodusPoint Capital Management in Paris and describes deep neural networks for calibration of a quadratic rough Heston model, with applications to SPX/VIX volatility and hedging quantities. A 2024 peer-reviewed publication lists Zhang’s ExodusPoint affiliation and describes an LSTM trained on hundreds of liquid stocks to forecast next-day realized volatility. Zhang’s public profile lists related topics including model calibration, volatility formation, news screening, and multi-asset market making. These records establish a dated researcher/publication trail. They do not establish that the papers were production ExodusPoint systems, that the exact methods were used by the firm, that Zhang remains in the same role, or that the work generated investment returns. See the capture note.

Evidence Boundaries

Podcast or video evidence is direct evidence of what a speaker or publisher presented. It is not independent verification of the claim. Vendor-hosted sessions can be customer-reported, sponsor-framed, or selectively edited. Job titles and historical interviews are date-scoped. Automated captions are navigation aids, not automatically exact quotations. No item here proves alpha, a live trading permission, comparative model performance, or an agent’s authority to allocate capital.

When the search can be called exhausted

There is no defensible universal proof that no podcast remains undiscovered. The practical stop rule is coverage exhaustion relative to a declared universe:

1. Every tracked firm, person, platform, language, geography, and source type has a recorded query and a status, including explicit negative results.
2. Every positive lead has been resolved to a canonical publisher, firm, conference, academic, or personal page, or is labeled unresolved with a reason such as paywall, deletion, login, or missing transcript.
3. At least three independent discovery routes have been run for each priority lane: feed/content search, guest/firm/role search, and conference/vendor or regional expansion.
4. Two consecutive sweeps produce no new canonical source above the promotion threshold in that lane. This is a diminishing-returns test, not a claim that the wider internet is empty.
5. The queue records what was not captured. A missing audio file, a blocked platform, a title-blind lead, and a reviewed negative result remain different states.

The current program has not met that bar. The Hedgineer RSS audit now resolves 42 episode enclosures, but only a subset has recovered transcripts; the episode inventory records the remainder as audio-located or publisher-summary-only rather than silently treating them as searched. The ledger is still English-heavy, the Singapore episode needs an archived audio/transcript, the ICML item is an official abstract without a confirmed recording, and several regional, vendor, researcher, and title-blind leads remain queued. The next meaningful milestone is therefore not “no podcasts exist”; it is two clean, independent passes with no new verified sources in each covered lane.

Source ledger: sources/06-industry-verticals/hedge-fund-ai-podcast-expanded-2026-08-16-raw.md; sources/06-industry-verticals/hedge-fund-ai-podcast-practitioner-signals-2026-08-16-raw.md; sources/13-multimodal-sources/podcast-discovery-missed-2026-08-18-raw.md; sources/13-multimodal-sources/blushing-quants-title-blind-expansion-2026-raw.md; sources/13-multimodal-sources/alpha-intelligence-title-blind-expansion-2026-08-18-raw.md; sources/13-multimodal-sources/flirting-with-models-index-crawl-2026-08-18-raw.md; sources/13-multimodal-sources/title-blind-institutional-quant-media-2026-08-18-raw.md; sources/13-multimodal-sources/podcastindex-watchlist-2026.json; sources/13-multimodal-sources/hedgineer-rss-inventory-2026-08-18-raw.md; sources/13-multimodal-sources/hedgineer-episode-captures-2026-raw.md; sources/13-multimodal-sources/podcast-transcripts-raw/.

Additional capture ledger: sources/13-multimodal-sources/cfm-philip-seager-aima-ep127-2026-08-18-raw.md; sources/13-multimodal-sources/ben-wellington-two-sigma-flirting-with-models-2026-08-18-raw.md; sources/13-multimodal-sources/stacie-mintz-pgim-flirting-with-models-2026-08-18-raw.md; sources/13-multimodal-sources/leda-braga-feg-insight-bridge-2026-08-18-raw.md; sources/13-multimodal-sources/balyasny-gappy-paleologo-ace-2026-08-18-raw.md; sources/13-multimodal-sources/hedge-fund-huddle-man-greg-bond-2026-08-18-raw.md; sources/13-multimodal-sources/hedge-fund-huddle-point72-academy-2026-08-18-raw.md; sources/13-multimodal-sources/aspect-capital-martin-lueck-jpmorgan-2026-08-18-raw.md; sources/13-multimodal-sources/bob-elliott-unlimited-hfnd-thor-2026-08-18-raw.md; sources/13-multimodal-sources/omer-seider-two-sigma-odds-on-open-2026-07-02-raw.md; sources/13-multimodal-sources/quantedge-ai-boundary-2026-07-30-raw.md; sources/13-multimodal-sources/title-blind-institutional-regional-expansion-2026-08-18-raw.md; sources/13-multimodal-sources/conference-title-blind-expansion-2026-08-18-raw.md; sources/13-multimodal-sources/future-alpha-2026-post-event-coverage-2026-08-18-raw.md; sources/13-multimodal-sources/title-blind-quant-researcher-media-expansion-2026-09-02-raw.md.

Latest regional capture note: sources/13-multimodal-sources/pendal-elise-mckay-ai-asx-2026-raw.md.

Verification status: canonical publisher pages and available transcripts were reviewed for the promoted items. Video captions remain navigation aids; current employment, model ownership, production permissions, and performance claims remain separately unverified unless a primary artifact says otherwise.

September 2 title-blind podcast routes: Voloridge and Quest

The Vitality Vision episode adds a markets-and-health podcast route naming Voloridge’s David Vogel, Barry Miller, and Daniel Hammack. The publisher page is dated March 9, 2026 and exposes transcript excerpts about Hammack’s progression from an internship and data-cleaning work into machine-learning and strategy responsibilities. Its public MP3 enclosure was recovered and transcribed locally into 1,096 timestamped ASR segments; the JSON, TXT, SRT, TSV, and VTT sidecars are durably retained in the private vault and private Hugging Face checkpoint. The ASR also records a speaker-reported statement that Voloridge was actively using LLMs for coding velocity, clearly naming Codex and Copilot; a third product was rendered as “Cloud Code” in two ASR passes and is left unnormalized pending human audio confirmation (approximately 00:25:06–00:25:21). The same discussion describes separate alpha, risk, volume/trading-cost, and cross-asset-correlation models (approximately 00:31:21–00:32:11). These are speaker-reported process signals, not proof of a firm-wide model inventory, agent permission, investment performance, or any particular product beyond the clearly transcribed names. Spotify and Apple Podcasts provide cross-platform locators. Full transcript bodies are not published. See the capture note.

The Forward Guidance episode with Mike Harris, dated August 19, 2024, contributes a Quest Partners route found through a generic quant-trading search. The publisher’s timestamp markers include CTA trend following, statistical arbitrage, multi-strategy trading, AI trading models, and the risks of training models with AI. The associated YouTube recording and Apple listing create additional recovery routes. No transcript is claimed here, and the metadata does not establish Quest’s current model inventory, data sources, permissions, production deployment, or performance.

September 2 title-blind audio-adjacent and practitioner routes

The expanded search found several routes that do not present as “hedge-fund AI” podcasts. The Confluent podcast with Ben Ford describes a former algorithmic-trading and hedge-fund practitioner discussing event-driven systems and operational change; the Kafka Summit presentation adds a related infrastructure and observability route. John Brennan’s operations podcast contributes a former Highfields Capital CTO perspective on digital transformation and organizational change. These are practitioner and infrastructure signals, not evidence of a named firm’s current AI stack.

Regional searches also recovered the Fidelity Canada episode with Gilbert Haddad and the AWS Singapore FSI Symposium programme, which names Lion Global Investors’ Head of AI of Investments and an AI/ML strategy track. These pages widen the media and personnel graph beyond English hedge-fund titles, but no complete transcript or firm-specific model disclosure is claimed. The expanded regional/vendor source note records the discovery boundaries and queued follow-ups.

The regional podcast sweep also found the SPH Money FM archive route, indexed with Syfe quantitative-research and portfolio-construction personnel, the German KPMG/Union Investment episode on human-controlled AI price validation, and the Indian Capitalmind Apple listing, whose metadata includes a chapter titled “How We Use AI at Capitalmind.” The first two are archive or publisher routes without complete public transcripts in this pass; the Capitalmind catalog does not expose the underlying workflow. These remain discovery records until the audio or a first-party technical artifact supports stronger claims. See the regional source note.

September 2 title-blind additions: Waters, Point72 alumni, and Two Sigma technology leadership

The Waters Wavelength SoundCloud catalog surfaced two generic capital-markets episodes that ordinary hedge-fund keyword searches can miss. Episode 347 with Brennan Carley, published March 5, 2026, lists agentic AI in capital markets, data-provider economics, data exclusivity, hidden implementation risks, and changing entry-level work among its topics. Its public SoundCloud audio was recovered and locally transcribed into a private timestamped navigation artifact. This is industry-technology context, not evidence of a named hedge fund’s production system or of Proton Advisors deployment. Episode 344, published January 13, 2026, adds a separate catalog route whose description names AI, agentic AI, quantum computing, and market data; no transcript or firm adoption is inferred from that description. See the capture note.

The AI Street interview with Kirk McKeown and the Podwise episode add a cross-platform person-first route for former Tudor, Glenview, SAC/Point72, and Point72 proprietary-research context. AI Street attributes to McKeown a transition from a high-volume manual research “factory” toward Carbon Arc’s knowledge-graph, ontology, and consumption-based data platform. The page also carries speaker-reported scale claims about structured transactions, data assets, storage, compute, and customer growth. These claims are useful clues about data architecture and research-process design, but they are not independently audited and do not establish that Point72 used Carbon Arc or that any current Point72 system has been disclosed. Historical and current affiliations remain separate.

The Canopy interview with Camille Fournier provides a different kind of evidence: a dated 2019 publisher transcript identifying Fournier as a Two Sigma Managing Director and former Rent the Runway CTO. The discussion concerns hiring, trust, team effectiveness, and indirect influence at senior levels. It does not disclose an AI model, data pipeline, or GenAI program. The linked MP3 returned 404, but the publisher HTML and normalized page text were retained privately; because the page has no audio timestamps, it is a searchable organizational-context artifact rather than a timestamped quotation source.

September 2 title-blind catalog expansion: Waters Wavelength

The Waters Wavelength SoundCloud catalog and its SoundCloud RSS feed add a new publisher-level discovery surface with 357 episode entries. This matters because the catalog is organized around capital-markets technology rather than hedge-fund or AI keywords. It exposes named guests, generic technology titles, and episode descriptions with chapter timestamps that can be searched before deciding which recordings merit recovery. The catalog inventory and episode-level records are in the media ledger.

Four episodes were recovered into the private transcript lane. Episode 352, published May 22, 2026, discusses vendor agent studios, vectorization, data quality, AI training, and governance. Its timestamped transcript includes a discussion of Symphony’s agent-studio announcement, FactSet’s Chief AI Officer reorganization and vectorization service, and Goldman Sachs’ AI bootcamps, Legend data platform, lineage, and digital-agent framing. These are reported media developments, not proof of hedge-fund adoption. Episode 350, published April 17, 2026, returns to data governance, lineage, partnerships, and scale as prerequisites for AI value. Episode 354, published June 19, 2026, contains unnamed bank-engineering commentary on vibe/agentic development and institutional bottlenecks. Episode 330, published September 5, 2025, was also captured; unsupported survey claims were excluded from the public synthesis.

The older catalog adds a useful historical vocabulary map. Episode 271 covers specialized language models and capital-markets use cases. Episode 243 identifies LSEG Labs and a deep-learning discussion. Episode 117 covers machine learning, chatbots, and innovation labs. Episode 86 names a Nasdaq surveillance leader and gives a dated machine-learning route. Episode 75 links Julia to finance research and risk modelling while discussing AI and alternative data. Episode 71 covers social data, sentiment, volatility, and event detection. Episode 70 describes OTAS analytics and AI-related buy-side decision support. These are date-scoped publisher descriptions; they do not establish current hedge-fund systems or investment performance.

The new media evidence sharpens the research question. Across the captured 2026 episodes, the disclosed implementation layers are data lineage, standardized context, vectorization, human review, developer workflow, and agent permissions. The recordings do not disclose model weights, training corpora, evaluation splits, portfolio authority, or reproducible returns. The capture note records the private checkpoints and the distinction between searchable transcript retention and public publication.

September 2 P0 personnel sweep: Numerai, asset-management data, and BlackRock Systematic

A P0 sweep over named people and firms produced three relevant recording routes after homonym screening. The results add research-process and data-infrastructure context; they do not justify a cross-firm ranking.

The official Numerai Office Hours S01E06, uploaded September 6, 2020, was not present in the curated ledger under its canonical URL. The public description and recovered captions expose a detailed discussion of live-versus-validation evaluation, model correlation, staking, and model diagnostics. Participants discuss how long a model needs to be observed before evaluation (23:56–25:33), whether live Sharpe versus validation Sharpe can diagnose overfitting (46:33–47:12), and why live-versus-validation differences cannot be cleanly assigned to overfitting, market regimes, model preferences, or feature exposure from a small validation slice (47:32–48:05). This is community and tournament evidence, not a current staff disclosure or a conventional fund’s internal model inventory. The official written recap provides a second identity and topic surface. The full timestamped capture is private in checkpoint 8200b117.

The person-index pass also recovered Epicenter 566 — Numerai & Predictoor, a September 2024 interview with Richard Craib and Trent McConaghy. It adds a dated bridge between Numerai’s equity-prediction feed and Ocean/Predictoor’s shorter-horizon prediction-feed architecture. The automatic captions support a bounded discussion of obfuscated data and modeling without semantic access to feature or stock identities (06:06–07:28), daily equity-prediction streams and the private/customer relationship around Numerai’s feed (12:59–14:23; 65:01–65:28), and a model range for prediction bots that includes linear models, boosted trees, Gaussian processes, neural networks, and transformers (43:33–44:09). The speakers also discuss prediction feeds as inputs for AI systems beyond trading (75:34–76:00). The date is deliberately recorded at month level because YouTube metadata and the guest’s talks index disagree on the precise interview/publication date. These are historical speaker statements, not evidence of current deployment, a current commercial partnership, model weights, data rights, or investment authority. The complete timestamped capture is private in checkpoint 31233094.

The Future of Finance panel on the data opportunity in asset management is dated January 19, 2021 on the official event page. Its named panelists are Greg Glass of Alpha Data Solutions, John Plansky of State Street, Jonathan Hammond of Sionic, and Ian Hunt of FundAdminChain. The recovered recording describes a data-management layer that reaches portfolio management, execution, risk, compliance, operations, and client reporting; it also discusses the use of reports, social, satellite, and mobile data in quantitative and AI workflows. Later discussion distinguishes a data model and on-demand views from a single warehouse (58:45–59:03; 1:03:17–1:03:41) and links cleaner data to automation and cost reduction (58:45–59:03). These are panelist and publisher statements about institutional asset-management infrastructure. They are not evidence that a named hedge fund deployed these designs. The private archive checkpoint preserves the full caption capture.

The short QuantMinds TV clip with Simon Weinberger, uploaded May 14, 2018, identifies Weinberger as Managing Director of Systematic Active Equities at BlackRock. Its title does not contain “AI” or “hedge fund”; the description frames the question around new data and the future of quant equities. The recovered caption is only 841 characters, so it supports topic and identity navigation rather than a full interview analysis. Its opening frames the operational requirement as finding new information sources, processing large volumes of data, and maintaining infrastructure and data-science expertise (00:09–00:52). It does not disclose BlackRock’s current model set, data contracts, training process, permissions, production status, or performance.

The three routes illustrate why the discovery system must retain broad finance and institutional-media searches even when an episode title lacks AI language. They add dated evidence about validation design, data architecture, and quantitative research infrastructure; they do not establish current firmwide AI strategy or comparable capability.

September 2 expansion: data vendors, regional managers, and research lineage

The title-blind pass added the Valsys Selling Signals catalogue, whose episode pages expose discussions of point-in-time integrity, entity resolution, dataset procurement, data quality, agent/MCP workflows, and the boundary between training data and tool input. Twelve publisher VTT transcripts are now retained in the private transcript lane and the verified private checkpoint. Five older audio-only feed entries remain unresolved because their audio enclosures returned HTTP 403; that is an access status, not evidence that no transcript exists. See the capture note.

The pass also added regional implementation accounts: Sumitomo Mitsui DS Asset Management’s AIR description combines multi-cloud GenAI, RAG, and specialized agents; BNP Paribas AM Japan describes AI-assisted patent-corpus analysis while retaining human stock-selection decisions; Shinhan reporting describes a rules/LLM hybrid orchestrator and internal agents; and regional reports add JoinQuant, Lingjun, Shanghai Minority, Bayes Capital, VanEck Australia, Guardian Capital, and RBC routes. Their evidence classes differ and remain separated: first-party implementation statement, job intent, manager-reported account, or secondary reporting. None supplies a complete model inventory, permission map, or independently verified performance.

The firm-controlled pass adds Acadian systematic-credit and ESG disclosures, CFM alternative-data/ML/agentic-AI and risk roles, D. E. Shaw market-data and alternative-data roles, GMO’s future quantitative-technology rotation, Jump’s ICLR and fellowship routes, Point72/Cubist data and quant roles, and Two Sigma’s data-as-code engineering account. These sources are useful because they surface governance, lineage, release gates, vendor controls, data reliability, and production-testing language that ordinary AI-title searches miss. They remain hiring or firm-description evidence, not proof of current deployment. The detailed source note is here.

The academic-personnel pass adds public routes for Two Sigma, Citadel Securities, Jump Trading, Tower, G-Research, Millennium, and a documented Man AHL internship. The linked researchers’ papers cover efficient transformers, LLM reasoning, generative modelling, diffusion, graph and sequence learning, neural operators, mean-field trading, limit-order-book reinforcement learning, meta-learning, and Bayesian policy learning. Those papers expose research lineages and possible search vocabulary; they do not establish that the employer deployed the method or that it contributed to returns.

September 2 title-blind expansion: recordings and regional operating clues

The title-blind recording note adds seven distinct media routes that would be missed by a hedge-fund, quant, or AI title filter. They include a CFA Institute panel with State Street, Fulcrum Asset Management, and Alphidence Capital participants; a Pete Muller/PDT Partners interview; two historical Two Sigma talks; a pre-GMO Christopher Heelan technical interview; a PanAgora/RavenPack panel; and a Bridgewater-to-Domino data-platform lineage interview. The available VTT and HTML transcript routes are marked for private retention; audio-only routes remain capture work. Historical affiliation and general data-platform discussion are kept separate from current fund systems.

The regional expansion note adds public signals from Schroders, First Private, Reimann Investors, Ontario Teachers’, HESTA, Evolved Reasoning, Eastspring, Temasek, ASIFMA, Annum/Turoid, Mizuho, Blackboard Asset Management, PL Asset Management, Astella, Absolute Investimentos, and Chinese industry and manager sources. Relevant clues include archive-based investment research, news filtering, systematic/data science teams, reporting automation, agentic-workflow hiring, and dated statements that some forecasting models were not yet used for investment decisions. These are not interchangeable evidence: they range from official pages and minutes to recruiting intent and regional reporting.

The personnel and lineage note adds HRT, Jump, and Cubist personnel surfaces with Harvard, Stanford, UC Davis, Imperial, Wisconsin, IIT Bombay, and other academic links. Their public topics include neural-network feature emergence, AI/ML algorithms, graph learning, compute platforms, kernel methods, reinforcement learning, bandits, and predictive-model validation. These links are useful for finding researchers, labs, and follow-on papers. They do not prove that an academic method moved into a firm’s production stack or generated an investment result.

September 2 publisher-transcript recovery: Invest with AI

The registered Invest with AI feed yielded five additional title-blind routes after the catalogue audit: Wall Street work mechanisms, Intelligent Alpha, Daloopa, Hudson Labs, and the Fable workflow discussion. All five publisher VTT transcripts are retained in the raw capture note and the private archive checkpoint 5c794492; the transcript bodies are not public.

The reusable signals are implementation questions rather than model-brand claims: moving from answer engines to work mechanisms; separating data coverage from workflow design; using structured paths, preprocessing, and checks around financial numbers; and testing models against long, context-rich financial tasks. Doug Clinton, Thomas Li, and Kris Bennatti speak for their own organizations or products. Their statements are not independent evidence of customer deployment, model performance, data permissions, portfolio authority, or returns. The audio files returned HTTP 403 in this environment, but the publisher transcript routes were durable and sufficient for private search and timestamp navigation.

September 2 title-blind allocator and manager media expansion

The allocator-media capture note adds twelve dated routes that ordinary hedge-fund, quant, and AI title filters would miss. They include publisher transcripts for Mike Pyle of BlackRock, Seth Klarman of Baupost, Ankur Crawford of Alger, Matt Cherwin of Marek Capital, and Chris Davis of Davis Advisors. Additional audio or catalogue routes cover Capital Group, PIMCO, Goldman Sachs, GQG Partners, Eagle Capital, Discovery Capital, and Titan Advisors.

The evidence classes vary: publisher HTML transcript, first-party appearance confirmation, audio metadata, premium-gated transcript, or rate-limited video route. The five transcript-bearing routes should be retained privately before any detailed synthesis. The remaining records are capture leads. A manager appearance or portfolio title does not establish an AI program, model, training data, deployment, portfolio authority, or performance result.

September 2 registry-wide podcast gap audit

The registry coverage audit checked all 74 registered shows without changing their seen-state. Thirty-seven RSS feeds responded and 37 registered surfaces were catalogue-only. The feeds produced 6,237 unseen entries matching their configured title/description discovery filters: 1,839 exposed publisher transcript URLs and 4,398 were audio-only. These are acquisition candidates, not 6,237 verified sources or 6,237 firm disclosures. The audit now gives the mining queue an explicit route for publisher transcripts before audio, and an explicit retry state for audio or catalogue items that remain uncaptured.

September 2 regional title-blind manager routes

The regional route note adds thirteen records across Japan, mainland China, South Korea, France, Germany, Brazil, Spain, Saudi Arabia, India, Australia, Canada, and the United Kingdom. The source classes include first-party manager pages, local financial reporting, a fund page, a job listing, and a Canadian firm podcast page. The signals range from quant-research descriptions and unstructured-data conversion to AI-enabled research engines and asset-owner AI-exposure mapping. Translation and local reporting risks are recorded explicitly; no item is treated as proof of a live model, deployment, authority, or performance.

September 2 title-blind podcast and event routes

The podcast and event route note adds twelve public routes found through generic allocator, factor, private- markets, macro, data-governance, and fixed-income searches. The set includes TDAM quantitative-investing leadership; CFA Institute events involving T. Rowe Price, Pictet, and Sparkline; Barclays quantitative macro; an official MSCI private-markets transcript; Fidelity Canada factor research; J. Rothschild, Lakeview Capital, Dimensional, Pictet, and AllianceBernstein appearances.

The records are intentionally not treated as interchangeable. Five publisher pages expose transcript text or a transcript PDF; the remaining items are audio, a member-only event, catalogue metadata, or an upcoming event. Six publisher artifacts were retained in private checkpoint bba60f11. The temporary fetch directory was staging only. No route establishes a firm’s AI deployment, model ownership, data rights, portfolio authority, or performance.

September 2 title-blind personnel routes

The GMO/CFM/Bridgewater personnel note adds dated public media routes for CFM’s Philip Seager; GMO’s Catherine LeGraw, Ben Inker, and Jeremy Grantham; and Bridgewater’s Karen Karniol- Tambour and Greg Jensen. It also records Arvind Thiagarajan’s current Tessel Biosciences role alongside a publisher-described former D.E. Shaw affiliation.

The value of this pass is temporal and personnel coverage. The interviews and episode pages do not, by themselves, disclose current AI systems, model ownership, training data, permissions, portfolio authority, or performance. The Tessel discussion is kept separate from D.E. Shaw; a former employee’s current biomedical work is not evidence about a former employer’s methods.

September 2 institutional and APAC comparator routes

The institutional/APAC route note adds public routes for GIC, CPP Investments, OMERS, IFM Investors, Amova/Nikko, Eastspring, Amundi, BTG Pactual, and AIMA Singapore. Relevant signals include enterprise AI assistants, internal research agents, quantitative and graph-learning research, earnings-call NLP, macro-factor estimation, and job descriptions naming RAG, vector databases, multi-agent frameworks, and cloud model providers.

Most of these are pension, sovereign, infrastructure, bank, asset-manager, or industry-association comparators rather than hedge funds. They are useful for context and discovery vocabulary, but their public pages do not establish a tracked hedge fund’s implementation, model ownership, data rights, authority, or investment outcome.

September 2 Two Sigma first-party recheck

The Two Sigma first-party route note adds five canonical pages that were visible in related research notes but missing from the central media ledger: Ben Wellington’s firm recap of Flirting with Models, his article on feature research, the New Seekers profile of MIT-trained quantitative researcher Abe Ejilemele, and Heather Miller’s ACM CAIS preview, plus Two Sigma’s January 2026 AI outlook roundtable.

Together they provide current title and personnel evidence, firm-authored language around shared feature libraries and orthogonal predictions, and a public description of agent-system evaluation, security, optimization, and operations concerns, plus a role-by-role 2026 outlook on internal context, feature generation, and overfitting risk. They do not establish a complete model inventory, training data, production permissions, portfolio authority, or performance. The five HTML pages are retained privately in checkpoint 0ef7b4e9.

September 2 Bridgewater and CFM first-party recheck

The Bridgewater/CFM route note adds a current personnel and lab-description lane. Bridgewater’s Nina Lozinski profile reports that AIA Labs grew from six people in 2023 to more than 50 scientists, engineers, and investors, while a current Greg Jensen profile connects his Managing CIO remit to AIA Labs. A separate Bridgewater article identifies Suri Bandler as an Architect on the Technology Team. CFM’s October 2025 publication describes its ML Lab as sitting within research teams and bridging current AI/ML work to academia.

The personnel titles and headcount are date-scoped, firm-reported evidence; the CFM statement is publication text hosted by CFM and is not an independent audit. These routes do not disclose complete rosters, model weights, training corpora, permissions, deployment status, or performance. The four captured artifacts are retained privately in checkpoint 56b61379.

September 2 underexplored low-coverage firm routes

The underexplored-firm route note adds five canonical pages found by starting with firms having relatively few ledger records. D. E. Shaw’s 2023 Machine Teaching publication describes optimizers in systematic and discretionary processes and iterative interaction between human judgment and machine outputs. Arrowstreet’s current What We Do page exposes technology-team, advanced-computing, research, optimization, and real-time-signal language without naming an AI model. Winton’s 2017 first-party AI essay is a historical discussion of non-stationary markets and technology claims. Aspect’s Anthony Todd podcast page and Dimensional’s Savina Rizova podcast page add named executive and research-leader routes.

The D. E. Shaw PDF is retained privately in checkpoint b18e34ca. The other four routes are first-party pages with metadata or firm-authored summaries; no transcript body was captured in this pass. These additions are coverage and chronology evidence, not proof of current AI deployment, model ownership, data rights, portfolio authority, or performance.

The second low-coverage pass adds current PDT work and careers routes, where PDT describes peer-reviewed quantitative research moving into live automated trading and lists Applied ML Scientist, Quantitative Researcher, and Research Engineer roles. It adds a current Systematica firm page that separates Research, Technology, and Trading and describes alternative-data and proprietary-technology work. It also adds Aspect’s Christopher Reeve podcast page and Martin Lueck/J.P. Morgan page. The latter’s publisher audio was recovered and privately archived; automatic ASR navigation places the discussion of hypothesis testing, hidden relationships, constrained datasets, signal/volatility/portfolio/execution applications, and LLM error risk at 15:32–21:46. Those timestamps require audio spot-checking and are attributed practitioner testimony, not a complete Aspect model or deployment inventory.

Two additional title-blind hiring routes are now recorded: Jain Global’s Singapore AI Research Intern listing, which describes LLM/agent evaluation and unstructured-data workflows, and a Harvard-hosted Old Mission graduate-trader listing, which exposes a university recruiting route and proprietary-trading description. Both are role or hiring evidence; neither establishes a filled AI role, live deployment, or investment result. Details and archive provenance are in the route note.

September 2 title-blind India quant-investing podcast recovery

The Success Documented interview with Rishi Kohli was not a conventional “AI hedge fund” title, but its chapter list contains a dedicated AI/ML-in-quantitative-finance segment. The publisher describes Kohli’s current Managing Partner and CIO role for Hedge Fund Strategies at InCred Alternatives and his prior ProAlpha/Monsoon, Avendus, LAQSA, ICICI Securities, and SSKI route. The episode also covers systematic investing, options, sector rotation, volatility, and quant-team management. Kohli’s LinkedIn post links to the same episode.

The public Captivate enclosure was recovered directly into the private vault; WhisperX-MLX produced 1,124 timestamped English segments from 64:54 of audio. The transcript and source audio are retained privately under checkpoint 03128af1. The ASR is for navigation and topic discovery, not a manually verified quotation source. The episode and metadata do not establish InCred’s current model inventory, data rights, production permissions, or performance.

September 2 Australian research-house perspective

The Zenith Investment Researcher’s episode is an adjacent route because its speakers report on meetings across approximately 650 Australian and global equity funds and ETFs. Ethan Spiegel and Bradley Antman describe six manager-reported use categories: operational efficiency, information sourcing, thesis challenge, industry research, data processing, and coding for quantitative managers. They also report an 86% historical observation for AI use in stock and industry research. The episode does not publish the underlying questionnaire, denominator, or manager list, so this is a dated research-house observation rather than a representative industry statistic.

The quant discussion distinguishes faster coding and data cleaning from a change in investment logic, while the broader conversation identifies IP leakage, internal hosting, cost, and agentic-tool adoption as practical constraints. The speakers characterize agentic workflows as a minority use case and describe Zenith’s own AI-assisted peer, attribution, regime, and outlier analysis. The public MP3 was recovered into the private vault and archived with 523 timestamped ASR cues in checkpoint 6538241a. The source note records the timestamped boundaries and makes no manager-ranking claim.

September 3, 2026 — RSS reconciliation recovers a Brazilian manager route

The current Sophron RSS feed exposed several quant and market episodes that were present in the publisher archive but absent from the individual capture queue. The most relevant new firm-linked route is the July 17 episode with Luciano Boudjoukian França, identified as a founding partner, CIO, and portfolio manager at Avantgarde Asset Management. Its description focuses on implementation rather than headline model claims: Brazilian factor backtests versus live-book costs, liquidity, point-in-time data, survivorship, corporate actions, capacity, and model override. It marks an AI-research-process segment at approximately 53:56.

Avantgarde’s first-party process page describes data cleaning and normalization, factor construction, security ranking, portfolio selection, risk management, and backtesting. The episode’s direct MP3 has been recovered into the private transcript-vault pending lane; the AI segment is not promoted as a verified quotation until the pre-existing single-GPU transcription jobs clear. The source note records the exact recovery status and the boundary between publisher metadata and firm-controlled evidence.

The Avantgarde team page adds a separate firm-controlled personnel and product surface: André Gouldbaum Lichtenstein is listed as the partner responsible for the AInvest Capital Inteligência Artificial Global FIA, described as a thematic fund focused on semiconductors, memory, energy, and data centers. The same page lists Brendo Henrique under Quantitative Research and describes UFAL economics and informatics training plus research in mathematical, econometric, and statistical methods. Avantgarde’s academic-engagement page names partnerships with Brazilian finance and university groups, while the AInvest site provides a fund-specific product and media surface led by Lichtenstein. These pages establish an AI-themed fund and named roles, but they do not disclose the fund’s model architecture, training data, research agents, or AI-attributed performance.

The same reconciliation adds a Yves Hilpisch episode about agentic coding, the backtest-to-live last mile, and quant-workforce changes, plus an Alphaparty episode with Rob de Rozario about digital-asset market structure and AI as a tool for small teams. These are quant-technology and adjacent digital-asset routes, not evidence of a tracked hedge fund’s model inventory or live trading authority. See the reconciliation note.

September 3, 2026 — Spanish podcast route on automation and data boundaries

The Spanish title-blind pass recovered an AzValor discussion with Director of Innovation Carlos Camps. A public transcript mirror provides timestamped windows in which AI output is treated as a first draft; public documents are used for initial summaries; client, prospect, and proprietary accounting data are kept outside cloud models; and an AI reconciliation model is tested in parallel with a traditional system. The transcript also describes human investment authority and a technical-talent pipeline. This is third-party transcript evidence pending audio review, not a model card or proof of production completion. See the regional route note.

September 3, 2026 — German TRYCON model disclosure

The German TiAM FundResearch interview with Michael Günther and Pablo Hess adds a named-system route that was absent from the earlier language pass. The interview describes Tungsten TRYCON’s proprietary QuantMatrix software, three model families and 48 AI systems operating in parallel, structured price and turnover inputs across approximately 60 markets, and an active-alpha component reported at roughly 5–10% of two public multi-asset funds. It also explicitly says the firm uses specialized internally developed models rather than large language models such as ChatGPT. Günther is identified with QuantMatrix development; Hess with AI-based and quantitative-strategy research and portfolio management.

This is manager-reported evidence about a named system, model organization, fund wrappers, and a stated human-controlled operating boundary. It does not publish weights, features, training windows, code, evaluation splits, order logs, approval rules, or the independent assessment mentioned in the interview. The source therefore expands the model taxonomy—it is not evidence that generative AI is used, nor an independently verified performance result. See the capture note.

September 3, 2026 — French Finaltis product route

The French Finaltis Cyril Systematic product page adds a first-party, title-blind product route. Finaltis describes a regulated UCITS absolute-return strategy trading equity, rates, and foreign-exchange futures, and says it incorporates more than 20 years of the firm’s quantitative research, including artificial intelligence. The page publishes share-class creation dates, ISINs, daily liquidity, leverage language, and VaR-based risk control.

This establishes a dated product wrapper and an AI-related research claim. It does not establish generative-AI use, model families, features, training data, retraining, execution authority, evaluation design, named AI personnel, or AI-attributed returns. The linked fund documents are the next verification route. See the capture note.

September 3, 2026 — French LFIS–QMI research-partnership route

The current French LFIS partnership page preserves a dated 2018 disclosure of LFIS’s partnership with the Quantitative Management Initiative. LFIS describes three research axes: AI and signal generation; portfolio construction and risk management; and implementation challenges. The page also identifies the Institut Louis Bachelier, Fondation du Risque, Paris-Dauphine, and ENSAE, and records a Quant Vision Summit held in Paris on October 4, 2018.

This is useful academic-partnership and conference evidence, but it is historical. It does not identify a model, dataset, model owner, current project, live strategy, client mandate, or AI-attributed performance. See the capture note.

September 3, 2026 — Spanish Pictet and Santalucía routes

The Spanish Pictet QuestAI article identifies Alexandra Nagy, Stéphane Daul, and Thibault Jaisson in the development lineage of QuestAI. Pictet describes approximately 400 features, including analyst earnings forecasts and preferences, boosting-tree models, quarterly retraining, and manager validation before trades. It also connects the strategy to a public research paper and Nagy’s EPFL/CERN background.

The Spanish Santalucía press release adds a current personnel route: Santiago Fernández joined the multi-asset team on September 2, 2026, with the firm describing his focus as AI applied to quantitative trading and process automation.

Pictet’s page is a first-party strategy and personnel account; Santalucía’s is a first-party appointment announcement. Neither establishes model weights, data rights, live logs, production ownership, or independently verified AI-attributed performance. See the capture note.

September 3, 2026 — Chinese routes expose agent, hiring, and live-integration language

The Chinese China Securities Journal interview with Lingjun CIO Ma Zhiyu adds a named-practitioner account of large-model use in quantitative research. The interview says Lingjun uses large models to process news, financial reports, research reports, and other unstructured or multimodal material for factor development; has deployed an investment-research agent for research-process iteration; and uses AI coding tools in risk-control development. It describes separate strategy, trading-system, and live-operation risk controls, while placing risk interpretation and investment judgment with people. Audio and video are described as future research inputs, not as an already disclosed production factor.

The Harvest Fund careers page provides a separate first-party hiring route. Its AI Analyst role names financial-large-model training, domain-specific fine-tuning, a global investment-decision engine, distributed training, quantitative deployment, and agent-system development. Other roles on the same page mention AI in fixed- income trading, quantitative risk and performance analysis, data engineering, cloud infrastructure, and security.

The Zhejiang University posting for Shanghai Xsquare Asset Management lists machine-learning research across price, fundamental, and alternative data, including Transformers, graph neural networks, reinforcement learning, factor mining, portfolio optimization, and transfer into live strategies. Its employer description says the firm develops its own mathematical models and trading systems across several asset classes and describes automated execution across multiple exchanges.

These three routes are different evidence types: a dated CIO interview, an official hiring specification, and a university-hosted employer posting. They do not establish filled roles, model ownership, training-data rights, live permissions, validation results, or independently verified AI-attributed performance. See the Chinese capture note.

September 3, 2026 — CFM’s French-language hiring surface specifies a modern model stack

The current CFM Machine Learning Researcher listing describes a newly formed quantitative-research team building AI-based alpha models. The listed work spans target definitions, data pipelines, model architectures, training recipes, distributed training, evaluation, and monitoring. Its model vocabulary includes mixture-of-experts, long-context Transformers, vision-language models, efficient attention, and multi-token prediction. It names PyTorch Distributed, DeepSpeed, FSDP, Megatron-LM, multi-GPU training, Ray or Kubernetes, CUDA, and custom kernels; preferred experience includes LLM-inference optimization and RLHF, RLAIF, DPO, or reward modelling.

A separate French interview with CFM vice president Lais Schunk dated March 4, 2025 describes alternative data, machine-learning and AI tools, price-pattern research, and an eight-person options and derivatives team. It also describes collaboration and method-sharing across CFM research groups.

The LinkedIn page is hiring-intent evidence; the interview is a dated practitioner account. Neither establishes a filled role, model owner, dataset or provider contract, production endpoint, capital authority, or AI-attributed performance. The two sources should remain separate rather than being collapsed into a claim that the advertised foundation-model stack is already deployed in a particular strategy. See the CFM capture note.

September 3, 2026 — Turkish and Polish product pages separate model support from fund authority

Aktif Portföy’s Turkish GLG fund page says asset selection and portfolio weights use outputs from the AI-based Magnus portfolio-optimization model. The displayed inputs and techniques include historical and peer prices, macroeconomic and company indicators, market and technical data, mean-variance optimization, Sharpe-oriented objectives, portfolio constraints, and covariance matrices. The page says portfolio-update suggestions are generated on period and threshold rules, with the portfolio manager able to adjust the cadence according to market conditions and manager views.

The related KAP record identifies GLG as a regulated qualified-investor fund, records a May 30, 2022 public-offering date, and lists Özlem Aydilek Arıcı as fund manager from June 12, 2026. KAP repeats the Magnus reference in the investment-strategy field.

The Polish Analizy.pl profile for UNIQA Selective Equity identifies Sebastian Liński as foreign-equity director and says he applies machine learning while managing equity funds and researching foreign companies. The same profile describes a separate bottom-up fundamental process and gives Liński’s SGH quantitative-economics, information-systems, and econometrics background.

These are product and personnel disclosures, not evidence of autonomous execution or model performance. The pages do not provide Magnus’s architecture, training data, validation, model version, or a complete manager-override map; the Polish profile does not identify a model or retraining process. See the Turkish and Polish capture note.

September 3, 2026 — Portuguese and Southeast Asian routes widen the evidence types

Copacabana Investimentos’ Portuguese first-party site names Daniel Cunha as fund manager and Marcelo Goldsztejn as Head of Compliance and Risk. Its Copacabana Quantitativo description says index-futures trade timing and sizing are determined by a proprietary algorithm. The page does not identify the model family, training data, validation, or execution controls, and does not claim GenAI.

The Portuguese-localized Goldman Sachs systematic-credit posting describes machine learning, TRACE/dealer-run/ECN data, predictive features, alpha models, KDB+/Q or ClickHouse, and an AI-driven self-service signal- backtesting platform with automated parameter tuning. This is bank recruiting evidence; it does not establish that every described system is deployed or name the team, model owner, or live results.

The MS Capital Singapore posting describes an AI/ML technology arm with deep-learning scientists, engineers, and quantitative researchers, and names market microstructure, fundamentals, events, multivariate data, model testing, and Python/C++. The Grasshopper Asset Management posting describes a Singapore quantitative-trading technology provider and an MAS- regulated licensed fund manager; its role mentions order-book and tick data, predictive models, econometrics, machine learning, simulation, data pipelines, and production deployment.

The Dynamic Technology Lab posting localized for Vietnam adds a Singapore-hedge-fund recruiting route for fundamental/quantitative research, Python/C++, and finance research. The official Schonfeld posting describes predictive signals, alternative data, backtesting, portfolio construction, execution, risk, NLP, unstructured data, and machine learning in a LATAM quantitative-research programme. These are recruiting scopes, not proof of filled roles or deployed systems.

An adjacent Indonesian route is the IPOT AI Trading announcement. It names AI Analytics, AI Trade Flow, AI Notification, AI Financial, and AI Real-Time Indicators and says the product processes tick prices, order books, broker activity, fundamentals, news, time series, pattern recognition, and probabilistic models. It is a broker/platform announcement, not evidence of an autonomous hedge-fund strategy. See the Portuguese and Southeast Asian capture note.

September 3, 2026 — Chinese filing and Thai prospectus add operating and personnel detail

The JF SmartInvest HKEX annual-results filing is a listed investment-advisory and financial-technology disclosure rather than a hedge-fund filing, but it is unusually specific about public AI products. It names FinSphere Agent Large Model Assistant V3.0, stock-diagnosis agent 4.0, an AI Monitoring Officer, and an AI Inspection Officer. The filing reports that V3.0 supported tool invocation, user-memory construction, and multimodal text-and-image responses; it also reports company-level customer, service, and token-usage figures. JF says it established a technology subsidiary and entered partnerships with Suntime, Tencent Cloud, and Nonconvex for financial-data integration, intelligent-advisory tools, cloud infrastructure, and AI-plus-quantitative services.

The same filing reports R&D spending, R&D headcount, intellectual-property and paper counts, but does not connect particular researchers to production components. The filing is company-reported evidence of product and organizational scope. It does not disclose provider contracts, training or retrieval corpora, data rights, evaluation fixtures, approval permissions, model-to-order paths, or independently verified investment attribution. The JF SmartInvest capture note keeps the filing separate from the company’s academic and product pages.

The SCB Machine Learning China All Share prospectus names a dedicated Machine Learning investment group with quantitative-equity responsibilities. Its personnel table identifies Dr. Poonsak Lohsunthorn with electrical engineering, mathematical finance, and mathematics degrees from the University of Southern California; Satitpong Chantrachirawong with computational-finance training at the University of Washington; Krit Jan-nak with financial-statistics and financial-engineering degrees from LSE and the University of Reading and prior roles at QIS Capital, JPMorgan Securities Hong Kong, and WorldQuant Research Thailand; and additional ML-investment staff with Chulalongkorn and Thai engineering backgrounds.

SCB’s current Thai Machine Learning Thai Equity product page says quantitative analysis and machine-learning techniques support security selection through a system developed by the manager. The prospectus and product page establish a named team, a regulated mutual-fund wrapper, and an explicit model-to-selection description. They do not establish the model family, training data, retraining schedule, individual ownership, production permissions, or AI-attributed performance. See the Thai capture note.

September 3, 2026 — Russian-language routes connect academic models to product disclosures

The HSE Artificial Intelligence in Mathematical Finance Lab describes multi-agent systems, reinforcement learning for market makers and hedging, generative models, agent-based market simulation, and reinforcement- learning delta hedging for illiquid markets. Petr Lukyanchenko is listed as lab head. HSE’s Financial Research and Data Analysis Center lists Tamara Teplova and Tatyana Sokolova and describes AI applications in stock and bond analysis, emerging markets, and commissioned projects. These are academic and center-level research disclosures; they do not establish a hedge-fund deployment, named industry partner, proprietary data source, or trading authority.

Alfa-Capital’s current AI strategy page says its strategy applies machine learning to Moscow Exchange stocks and depositary receipts, scores securities in real time, holds positions for hours or days, and uses more than 100 parameters per security over roughly two decades of Russian equity history. Alfa’s experts page names Nikita Elenberger as Portfolio Manager for Algorithmic Strategies and Ramazan Teshev as Head of AI Service Development. These pages do not identify the model family, feature definitions, labels, validation, data vendors, or which person owns the investment model. The June 2026 AKQU review adds a current product route but not model architecture or attribution.

The current Finam QuantPro product page describes more than 200 algorithms, automated long/short trading across futures, stocks, and options, risk limits, options hedging, and no human participation in individual trades. It labels the system algorithmic rather than specifically ML or GenAI. A Russian VTB podcast episode dated March 29, 2024 names VTB data-analysis/modeling leader Denis Surzhko and prospective-ML-algorithms leader Evgeny Lepshin. The podcast is adjacent bank evidence, not evidence of a hedge-fund system. See the Russian capture note.

September 3, 2026 — localized media routes expose titles and summaries that English searches miss

The Japanese DigitalCast page for Jim Simons links to the official TED recording and provides an English/Japanese script. The recording describes hiring mathematicians, designing algorithms, computer-testing models, collecting large datasets, and searching for persistent anomalies. This is a subtitled archive, not Japanese-language speech.

The Korean-distributed SamproTV Ray Dalio interview has a third-party timestamped transcript mirror. The route discusses Bridgewater’s historical study, country-level financial and geopolitical analysis, and independent thinking. The transcript should be checked against the audio before quotation.

The Chinese Apple Podcasts episode with Zhang Lei identifies Zhang with Hillhouse. Its show notes describe research across time, regions, industries, and online/offline sources, but no transcript was exposed.

The Spanish Fundación Rafael del Pino Ray Dalio conference has an official video page and Spanish audio. Its written summary describes converting decision criteria into equations and algorithms, observing results over time, and iterating through goals, failure, diagnosis, system design, and implementation. The summary is not a timestamped transcript.

The Russian TechFlow translation of Bridgewater’s 50th-anniversary recording links to the original recording and presents Ray Dalio and Jim Haskel discussing systematic testing, backtesting, learning from errors, and research culture. The page labels the text as a translation, not Russian-language speech.

Two additional English-language routes were found through the same title-blind expansion. Bridgewater’s OceanXplorer recording embeds a 2024 discussion of debt cycles, portfolio construction, geopolitics, climate, and possible AI effects on productivity and inequality. The World Economic Forum recording describes climate-finance and de-risking innovation rather than a trading workflow. The Rob Carver Personable episode dated May 30, 2026 identifies a former AHL fixed-income head and provides chapter markers for discussion of systematic execution and momentum. It does not establish current Man Group systems or AI use. See the foreign-language media capture note.

September 3, 2026 — Chinese recruiting and industry-standard routes add model-governance detail

The QP Alpha official profile identifies Shanghai QP Alpha Investment Management, gives an AMAC registration, and names Sun Lin as founder/CEO/head of investment research and Yu Hang as founder/CTO/ head of information technology. The biographies describe training at Fudan, Imperial College, Indiana, and Carnegie Mellon and prior roles at Barclays Capital, Two Sigma, Knight Capital, and Tower Research. QP Alpha’s strategy page names index-enhancement, market-neutral, transaction-cost, stock-selection, and CTA model categories; its English page states that machine learning and distributed computing are used in trading strategies. These are firm-published claims and do not disclose model inventory, training data, validation, or order authority.

A distinct Zhejiang University IFQuant posting specifies a data-and-machine-learning scientist role covering dataset validation, factor discovery, feature construction, ML/deep-learning prediction, live-strategy metrics, Python/C++, and Linux. It is hiring-intent evidence, not proof of a filled role or a deployed model.

The China Securities Investment Fund Association large-model application standard adds an industry-level governance route. It specifies provenance and versioning for models and datasets, license records, separate training/data/ invocation permissions, RAG traceability, factor mining, model simulation, compliance review, testing, and audit logging. It names contributing institutions and technology providers, but does not prove that any one firm has deployed every listed capability. See the Chinese IFQuant and QP Alpha capture note.

September 3, 2026 — Southeast Asian follow-up adds regulatory and product boundaries

The current SCB Machine Learning Thai Equity page and factsheet confirm the fund’s quantitative and machine-learning security-selection policy, describe the system as manager-developed, and name Poonsuk Lohsoonthorn and Pairit Nittayanuparp as managers. KTAM’s AI Brain RMF page describes another Thai-equity product with quantitative selection, periodic fundamental-data updates, daily optimization, and possible automated order routing; it says the algorithms may be developed by KTAM or another application developer. Neither route identifies a model family, training data, developer contract, validation design, or model permissions.

The MAS Grasshopper record confirms the Singapore CMS-licensed fund-manager entity and identifies Yap Lien Sern Gabriel as CEO and executive director. The MAS JCube record identifies Chan Sai Pang as CEO and adds a regulatory anchor to JCube’s already captured systematic/ML website. The WeInvest team page names Bhaskar Prabhakara, Elango Balusamy, Rajesh Arjunlal, and Chiranjeet; the MAS Planar record links the platform to a regulated Singapore entity. This is platform and regulatory evidence, not a confirmed hedge-fund model disclosure.

The Shen Yao / Plutus Mazu page describes a Singapore-incubated quantitative hedge fund using diversified signals and AI-refined predictive models, but does not provide a public model, regulatory record, named research team, or performance evidence. In Vietnam, the AlgoTrade page names Vo Duy Anh as CEO and Nguyen An Dan as Chief Scientist and describes human–AI hybrid trading; the Genquity site names leadership and an AI-engineering role and describes agents processing Vietnamese financial statements and macro conditions. These are first-party claims requiring independent confirmation of capital, licensing, employment, and results. See the Southeast Asia capture note.

September 3, 2026 — title-blind infrastructure roles expose the production boundary

Two Point72/Cubist roles and a second platform-engineering posting add Warsaw-based data-platform and delivery-control routes to the existing Cubist research record. They mention multitenant data, operational and historical data, Databricks, streaming, Unity Catalog, AWS, Spark, Delta Lake, Kubernetes, Terraform, CI/CD, automated security checks, and guardrails. The roles do not name a model, agent, RAG corpus, or data contract.

Schonfeld’s Platform Support Engineer posting describes 24/7 monitoring of a proprietary quantitative-trading platform, real-time incident response, change control, portfolio-manager support, code deployment, and recurring-process automation. Its Business Analytics role covers Python infrastructure, accounting, compensation, reporting, and minimal manual intervention. These are operational and non-investment automation signals, not evidence of AI-model deployment.

Man Group’s Senior Storage Automation Engineer role describes a three-engineer storage team, infrastructure-as-code, Azure, Python, Pure Storage, Rubrik, VAST, and 24-by-365 support, with research and development teams among the capacity and performance stakeholders. Balyasny’s database platform role mentions declarative provisioning, automated lifecycle management, self-healing operations, Snowflake, and Terraform. Neither posting establishes an AI model, investment authority, or performance. See the title-blind infrastructure capture note.

September 3, 2026 — previously unindexed manager media routes

A title-blind, cross-platform reconciliation recovered public transcripts and recordings for Acadian, CFM, Arrowstreet, Man, Bridgewater, Two Sigma, Citadel, and Renaissance. The routes add named personnel, founder history, academic context, and strategy discussion to the media graph even when an episode title contains none of “hedge fund,” “quant,” or “AI.”

Examples include John Chisholm’s Acadian transcript, Jean-Philippe Bouchaud’s CFM episode, John Y. Campbell’s Arrowstreet conversation, and Giuliana Bordigoni’s Man Trend Setters episode. Bridgewater routes include a GIC transcript, a Greg Jensen discussion, a former-employee ReSolve interview, and a Milken transcript with Ray Dalio. The same pass added a Two Sigma founder interview, a Citadel conference transcript, and two historical Renaissance routes.

These sources are discovery and context evidence. Public transcripts—especially automatic or third-party transcripts—must be checked against the audio before quotation. None establishes model weights, training data, data rights, current model ownership, live investment authority, or AI-attributed performance. See the media recovery capture note.

September 3, 2026 — Central European and Persian routes

Local-language searches added Czech and Slovak quant-research hiring and conference routes, Hungarian asset-management AI material, Bulgarian AI-fund and multilingual-finance benchmark pages, Romanian financial-ML research, and Persian portfolio-optimization research. The Qminers ML researcher opening mentions quantitative research, feature engineering, Python, and trading-oriented modelling; FinMMEval defines multilingual financial-LLM Buy/Hold/Sell tasks; and AI4EFin covers ML/AI for energy-market forecasting, causal analysis, and risk.

UBB’s Optimum Fund documents describe an AI-managed Bulgarian fund using a human–machine partnership, while K&H describes AI-supported allocation with human-controlled management. Russian routes add Finam’s AI Laboratory, which describes work from LLM benchmarking through trading and risk. These are product, firm, academic, or hiring disclosures—not proof of a common regional stack or live investment deployment. See the CEE/Persian capture note.

September 3, 2026 — additional title-blind personnel and research routes

The latest cross-platform pass adds public personnel and research routes for Marshall Wace, Squarepoint, Brevan Howard, Winton, Schonfeld, and Renaissance. For Marshall Wace, the AIMA report and Washington Post/Bloomberg interview are historical leadership commentary on quantitative hiring, quantamental capacity, and alternative data. OxWoCS and UZH add named quant and academic-placement routes. The public EconBERT paper and model page are academic artifacts; they do not establish use inside the firm.

Squarepoint routes include an IAQF Boston event, a Columbia employer session, an EPFL seminar, and self-reported personnel and open-source material from Jeevan Devaranjan. The event and recruiting pages use systematic-strategy, data-analysis, and compute-platform language. They do not disclose model ownership, permissions, or performance.

Brevan Howard’s Neudata biography attributes enterprise-data, alternative-data, and AI-solution responsibilities to Joseph Aube’s role; the AFI recap adds a named CTO event route. Winton routes include David Harding’s Goldman Sachs conversation and Alpha Lee’s TWIML episode. The former is founder commentary; the latter concerns academic and startup ML work associated with a fellowship. Neither establishes a current trading-model inventory.

Schonfeld’s Quantbot event recap and IMU report add technology-event routes involving named personnel. Academic biographies for former personnel remain separate from firm deployment evidence. The Simons Foundation interview and institutional profiles for Stephen Della Pietra add Renaissance historical computing and NLP lineage, not a current system disclosure.

September 3, 2026 — South Asian regulatory and product discovery

The SEBI AI-only AIF circular is a regulatory vocabulary route, not evidence that a named fund has migrated. A Morningstar account of Tata Quant Fund describes a historical product using rule engines, predictive models, factor inputs, and recalibration. ArthAlpha’s APMI page and Sixteen Alpha’s first-party page add current self-described Indian quant/AI product routes, but their model details, operating status, and independent validation remain unresolved.

In Malaysia, the Securities Commission DIGID catalogue lists AI and ML solution claims, while BFM’s investor-AI episode adds a publisher audio route. The IIT Bombay capital-markets centre provides a research route covering algorithmic finance, high-frequency trading, and AI/ML. Additional Bengali, Hindi, Marathi, Telugu, Malay, and Indonesian product and hiring routes are preserved in the South Asian source note.

September 3, 2026 — conference and vendor discovery routes

The AIMA Technology & Innovation Day speaker roster and AIMA Global Investor Forum agenda create forward monitoring routes for named technology, portfolio, and investment personnel. The NVIDIA GTC24 session with BlackRock describes financial knowledge graphs, document retrieval, and RAG in its abstract. A roster or session abstract establishes a public event route; it does not establish attendance, production deployment, model ownership, data rights, or investment impact.

The new routes expand the search graph and the follow-up queue. They do not support a cross-firm capability ranking. Full provenance and boundaries are in the major-firm route note.

September 3, 2026 — academic agentic-alpha route

The XALPHA paper describes a memory-driven AI quant researcher with separate macro, micro, and cross-cycle components. It connects report-grounded research memory to hypothesis generation, code, financial-plausibility checks, backtesting, and feedback across generations. The paper reports CSI300 experiments, but this is academic evidence rather than evidence of adoption by any tracked firm. The frontier-lab source note records the broader verification pass and its source boundaries.