Podcast Title-Blind Follow-Up: New Quant and Hedge-Fund AI Signals

Related research: Hedge-fund AI podcast and video practitioner signals · Finance and quant audio/video deep dive · External Media Research Room

Episode URL: N/A — this is a multi-episode discovery and transcript-capture note; individual recording URLs are cited inline.

Credibility: MEDIUM overall; HIGH for canonical publisher metadata and firm-controlled video locators, and lower for speaker or vendor descriptions that are not independently corroborated.

Source files: sources/13-multimodal-sources/podcast-title-blind-followup-2026-08-17-raw.md; research/13-multimodal-sources/{acadian,alphasimplex,citadel,bridgewater}/raw/.

What the follow-up added

The personnel sweep produced 241 YouTube candidates, most of which were irrelevant matches or duplicates. Transcript-content search and canonical-page review reduced that queue to a smaller set of useful signals. The new material falls into five distinct evidence lanes: alternative-data factor design, agentic research workflow, deterministic institutional infrastructure, options-specific ML, and methodological controls for predictive models.

This is a map of public disclosures and research vocabulary. It is not a ranking of firms, models, or people.

Newly processed signals

1. Kai Wu connects GMO lineage, intangible assets, and NLP

In The Acquirers Podcast’s September 18, 2025 episode, Kai Wu describes starting his career at GMO, working for Jeremy Grantham, participating in a GMO-linked quant-fund spin-out, and founding Sparkline in 2018. In the timestamped transcript, he describes machine learning, alternative data, and NLP as tools for quantifying intangible-asset pillars such as intellectual property, brands, human capital, and networks (approximately 01:25–08:40).

The research signal is specific: unstructured corporate information is being translated into structured factor inputs for an existing investment process. That is different from using a general-purpose language model as a stock selector. The source does not disclose model architecture, data licensing, portfolio permissions, or performance attribution.

2. A small-fund workflow makes the agent boundary explicit

Fundamental Edge’s June 26, 2026 interview with Raj Shah describes Shah as Stoic Point’s co-founder and a former Light Street partner. The publisher’s chapter list separates screening, research, and monitoring; it also identifies deterministic versus non-deterministic screening, computer use, automated monitoring, AI-assisted pitch review, and ROI measured in P&L rather than hours (04:00–35:59).

The useful implementation detail is the separation of deterministic filters from probabilistic assistance. The episode also describes a meta-screen, monitoring, and a research sparring partner. These are reported workflows from the conversation, not independently audited production evidence.

3. “Agentic quant” is being framed as orchestration plus controls

Denis Lukyanov’s March 27, 2026 Blushing Quants episode frames LLMs as orchestrators, analysts, and research accelerators. The publisher description explicitly pairs that with context systems, guardrails, model-as-judge workflows, regime detection, explicit logic, strong data, and human control.

That vocabulary is useful for source classification: an agent can sit around a predictive research stack without being the predictive model or the authority that allocates capital. The episode is not attributed to a named hedge fund.

4. Epoch exposes the infrastructure layer behind institutional AI

Garret Brennan’s April 27, 2026 episode identifies him as Epoch’s co-founder and CEO. The publisher describes Epoch as a structured AI interface over an institutional research stack, with determinism, hallucination risk, backtesting, internal libraries, and multi-agent workflows.

This adds a recurring implementation pattern to the research map: natural- language interaction is placed above deterministic computation and existing research infrastructure. It is vendor/startup evidence, not proof of a client deployment.

5. OneEye describes an options-specific ML stack

Eren Biri’s May 25, 2026 episode identifies Biri as founder of OneEye Capital and describes an in-house options stack, large options datasets on the firm’s own hardware, and AI/ML use for signal calibration, regime classification, portfolio optimization, and empirical pricing.

This is a useful small-manager architecture lead because it names several model-adjacent tasks without claiming that a language model forecasts returns. The details remain manager-reported and should be checked against firm pages, job descriptions, filings, or technical artifacts.

6. Factor research coverage added validation vocabulary

Antonio Marrazzo’s July 22, 2026 episode covers multiple-testing bias, realistic tradable universes, liquidity and transaction costs, point-in-time data, look-ahead bias, classification, logistic regression, random forests, XGBoost, purged validation, and embargo periods.

This is methodological context rather than a firm disclosure. It is valuable for reading public claims because a model name alone says little about whether the reported result survived timestamp alignment, costs, multiple testing, and out-of-sample validation.

7. A public methodology counterpoint separates predictive ML from GenAI

In David Wright’s Monetary Matters episode, the timestamped transcript distinguishes a predictive ML system from generative text models. The episode describes a tree-based aggregate model retrained quarterly for daily 20-day forecasts, while language-model projects are described for sentiment signals from earnings-call transcripts, analyst reports, and news. It also discusses temporal leakage and the need to align text training data with the information available at the decision time.

The practical classification is clear: generative AI can assist text transformation or research workflows, while a separate predictive model can consume structured or extracted signals. The episode is methodology evidence, not a named hedge-fund deployment or return claim.

YouTube additions captured through the browser path

The signed-in Chrome-cookie path successfully retrieved English auto-captions for the initial four videos and then for a channel-level crawl of The Fund AI Pod. The channel crawl is recorded in the source artifact. Captions are retained locally with timestamps and treated as navigation aids, not as clean public transcripts.

What remains in the queue

The remaining acquisition queue includes the remaining Fund AI Pod episodes and finance-specific channels. Carmen Li’s compute-futures episode, OneChronos’s compute-market interview, and Cliff Asness’s AQR discussion are now processed with separate evidence boundaries. The ex-Balyasny and Evolution episodes have now been recovered and moved into the primary-audio review ledger. DeepValueIntelligence has a dated indexed GitHub record, but direct clone/API/raw-file requests returned 404 on this pass, so it remains a reproducibility lead rather than a current codebase.

These are acquisition leads, not promoted firm findings. The queue is deliberately separated from reviewed evidence because metadata and transcript search can overstate relevance.

Evidence Boundaries

The source ledger records the capture method, timestamps, and boundaries: podcast-title-blind-followup-2026-08-17-raw.md. No item here establishes alpha, live trading permission, proprietary model performance, or current employment beyond what the linked source explicitly states. Full third-party transcripts and media are not reproduced on the public page.