Serafis Mines Podcast Transcripts So Hedge Funds Don't Have To

The Macro: The Best Alpha Is Hiding in Unstructured Audio

Institutional investing runs on information edge. Every hedge fund, every asset manager, every family office is trying to find signals before the market prices them in. For decades, that meant reading SEC filings, earnings transcripts, and sell-side research faster than everyone else. Bloomberg, FactSet, and Refinitiv built empires providing structured financial data. AlphaSense and Sentieo added natural language search on top of traditional document sets. That infrastructure is mature and well-covered.

But the information landscape has shifted. The most interesting signals increasingly live outside of traditional document formats. A fund manager drops a 90-minute interview on a podcast explaining their thesis on energy transition. A CEO gives a candid 45-minute talk at an industry conference that never makes it into a press release. A founder does a podcast tour discussing competitive dynamics in their market. This audio content is information-dense, often more candid than formal filings, and almost entirely unsearchable.

The scale of the problem is staggering. There are thousands of finance-relevant podcasts producing new episodes every week. Investment conferences generate hundreds of hours of recorded talks per quarter. Investor letters and interviews are scattered across dozens of platforms. No research team has the bandwidth to listen to all of it, and the traditional financial data providers have been slow to index audio content meaningfully.

Some firms have tried to solve this with in-house transcription and analysis teams. It is expensive and does not scale. Others use general-purpose transcription services and then manually search the output. That gives you text but not insight. Knowing that a fund manager mentioned a company name is not useful unless you can track how their conviction has changed over six months of interviews, cross-reference it against what other managers are saying, and filter by the credibility of the source.

AlphaSense has added some podcast transcript coverage, but their core product is built around SEC filings and broker research. Tegus focuses on expert network interviews, which is adjacent but different. Koyfin and Daloopa are great for financial data extraction but do not touch audio content. The dedicated infrastructure for mining investment-relevant audio at scale does not really exist yet.

The Micro: A Former Ribbit Capital Data Lead Building for His Own Frustration

Serafis was founded by Rohan Sharma. His background tells you everything about why this company exists. Rohan was Head of Data at Ribbit Capital, a fintech-focused fund managing $12 billion. Before that, he was a software engineer at Google. He also bootstrapped and exited Kosmos, a data integration platform for wealth managers. He studied computer science at Princeton.

This is a founder who has lived the problem from the inside. At a $12 billion fund, the research team is sophisticated and well-resourced. If even they struggled to systematically mine audio content for investment signals, every smaller fund is dealing with the same gap.

He came through Y Combinator’s Summer 2025 batch. The team is now four people in San Francisco. The product already serves 12 organizations managing over $70 billion in capital combined. For a company less than a year old, that is remarkable traction in a market where sales cycles to institutional investors are notoriously long.

The product is an indexed library of podcast transcripts from investors, founders, and corporate executives. Users can search by person, company, or theme. They can filter by the credibility and information density of the source. The knowledge graph layer is what elevates this beyond simple transcript search. It connects entities across conversations, tracks narrative evolution over time, and identifies when consensus is forming or shifting on a particular theme.

That last capability is the real product. In institutional investing, the money is made by identifying narrative shifts before they become mainstream consensus. If three respected fund managers mention the same thesis on different podcasts in the same month, that is a signal. If a CEO who was bullish on a market segment six months ago has turned cautious in recent interviews, that is a signal. Extracting these patterns from unstructured audio is something that humans do poorly at scale and machines can do well.

The Verdict

I think Serafis has found a genuinely underserved niche in a market that spends freely on information advantages. Institutional investors are not price-sensitive when it comes to research tools. They are accuracy-sensitive and speed-sensitive. If Serafis can surface a narrative shift 48 hours before it becomes sellside consensus, the product pays for itself on a single trade.

The founder-market fit is as clean as it gets. Rohan built data infrastructure at one of the most respected fintech funds in the world. He knows exactly what institutional research teams need, what they are willing to pay, and how they evaluate data products. That knowledge is worth more than a technical cofounder in this market.

In 30 days I want to see coverage depth. How many podcasts and audio sources is Serafis indexing? The value of a knowledge graph scales with completeness. If they are covering the top 50 finance podcasts but missing the long tail of industry-specific interviews, the product has gaps that sophisticated users will find quickly.

In 60 days the question is signal quality. Are the narrative shift alerts actually predictive, or are they just interesting? Institutional investors will pay for interesting insights temporarily, but they only keep paying for tools that improve their investment process measurably.

In 90 days I want to understand the competitive moat. Transcript indexing is a commodity capability. The knowledge graph, the credibility filtering, and the narrative tracking are the defensible parts. If a competitor can replicate the core value by pointing a general-purpose LLM at the same transcripts, Serafis needs a deeper moat. If the knowledge graph and entity resolution are genuinely hard to replicate, the 12-customer head start becomes a real advantage.

Seventy billion in managed capital across 12 clients in the first year is a strong signal. The institutional investor market is small in number but enormous in spend. If Serafis can become the default research tool for audio-derived investment signals, this is a company that could build a very profitable business without ever needing to be massive.