Creative LLM use cases - Podcast agent, investor letters, synthetic panels
Plus 2 interesting job descriptions and 1 new tool
Flat Circle tracks creative use cases of LLMs in hedge funds. If you haven’t already, join hundreds of PMs, analysts and engineers reading each week:
Interesting use cases
Have an LLM listen to every industry podcast and alert you with anything relevant using our new Podcast Agent
26 million podcast episodes are published each year. Many are interviews with management teams from public companies, their competitors, customers and suppliers. Based on interest from several funds, we built a simple tool that listens to every podcast and alerts you of anything incremental to your coverage. Couple examples from this week:
ORCL: VP of a 17 hospital system discusses their recent decision to transition off Oracle Health (Becker’s Healthcare Podcast)
COTY, ULTA: Ulta’s SVP of Ecom discusses their new marketplace, how it plans to bring emerging brands into their brick and mortal channels (Omni Talk Retail)
We plan to add more sources of content to this over time. Sign up for free.
Askelladden Capital discusses its process for ingesting and scoring fund LP letters
I’ve built a tool that reads investment letters of other fund managers, ignore all macro / philosophical discussion, extract only single-ticker investment ideas, summarize them, and scores them against a rubric based on our historical priorities. That rubric was – drumroll – drafted by AI after reading years of our letters, then subsequently refined by me.
My take: If you’re building this agent for your process, Yellowbrick is a pretty good datasource for investor theses.
Prompt from Reddit commenter to analyze 10Ks/Qs for changes in forward looking statements.
My take: Agree with the focus on using LLM to extract verbatim language vs return a conclusion.
Interesting papers
LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings. Surveys a panel of LLMs pretending to be shoppers and compares results against real panelists, finding a 90% overlap. Prompts discussed in the paper and on their github.
My take: There’s a lot of synthetic panel research vendors including Aaru, Electric Twin and Qualtrics. I think synthetic panels work almost as well as human panels, which is not that great. The best ones are calibrated against real panel and purchasing data. I’m not clear why a panel of LLMs is better than a system leveraging a single LLM, or even what the difference is. Synthetic panels are also unlikely to produce entirely new findings like humans. They also might be more useful for companies for feedback on a specific product, vs investors who are looking for fresh, on the ground information. There’s a lot of interest in this space, so will explore it more and report back with anything interesting.
New tools
SimilarWeb has started offering 12 months of data free through Manus paid and free plans
My take: SimilarWeb already offers an MCP server for its existing customers to access via LLM, the difference here is their move to share data for free as lead gen. I think we’ll see a lot more of this, especially in relatively commoditized segments like web traffic data. Worth noting this was announced two weeks after Meta acquired Manus for >$2b.
Interesting job descriptions
AlphaSense/Tegus is hiring a hedge fund LLM workflows product manager
your value lies in your ~5 years of experience at a top-tier Hedge Fund…You will be the primary architect of “quality.” Design, test, and refine prompts to ensure our AI output meets the high standards of a professional investor. You will look at an AI summary or extraction and immediately know if it “sounds right” to a PM or Analyst
Millennium hiring GenAI engineer for advanced RAG
Follow for more case studies of LLMs+Investment Research
If you would like to discuss incorporating LLMs into your research process, reply to this email or reach out via X or LinkedIn.