Fifty-four thousand executive interviews, made queryable

The situation

This started as a body of interviews to train on and a proposed AI interviewing tool. It evolved into something more useful: an executive-interview media-analysis platform.

What we did

The ingest layer discovers, downloads, and transcribes executive-interview podcasts from dozens of sources, using on-device transcription where possible to hold cost down. It runs continuously, so the corpus tracks what market leaders are saying now instead of freezing at the date it was built.

Processing runs a batched, multi-pass classification pipeline over the whole corpus: a three-level role × industry × scale taxonomy, then thematic and buyer-journey scoring on top of it. The output is a vector-searchable dataset delivered through a chat interface anyone can query in plain language, with industry, role, and company-scale breakdowns feeding media analysis and an ICP profile builder.

Why it matters

The interesting constraint here was economic, not technical. Anyone can classify a corpus this size with a frontier model and a large budget. Doing it at a cost that makes the dataset worth having is the real engineering problem, and it is the same problem every portfolio company hits when they move from pilot to production volume.

• Operating leadership · Multi-sector · Lower-middle-market portfolio

  Growing portfolio revenue from $10M to $180M

• M&A · Healthcare · Media · B2C · Revenue cycle management

  Buying and selling companies, and getting the deal closed

• Turnaround · Out-of-home media

  A million a month in losses, taken to break-even