Six free data sources, one LLM, and 3,872 AI companies
Zero paid APIs. The Wayback Machine, SEC Form D filings and four other public sources, reconciled into one dataset and classified by a single LLM pass.
- The catch
- Haiku failed the canary test on ambiguous company descriptions; Sonnet was the sweet spot on cost against accuracy.
- Takeaway
- Most "you need a paid data API" problems are really entity-resolution problems wearing a disguise.