Natural-language search · shipped in production

Search companies in plain English.
Watch it compile to a live query.

This is the natural-language search engine I designed and shipped in production at a fintech, rebuilt here as a demo you can type into. It knows ~40 well-known startups — ask by valuation, sector, funding stage, geography, or headcount. Every result reveals the typed query your English compiled to; that translation is the engineering on display.

It’s a live, playable prototype — not a package or product. Nothing to install or sign up for; just type and run.

  1. 1Plain English

    Ask about companies the way you'd ask an analyst — no filters, no query syntax.

  2. 2Compiled to a DSL

    An LLM parses it into a typed query — filters, sorts, limits — validated against a schema.

  3. 3Runs live

    The structured query executes over the dataset and returns matching companies in milliseconds.

Scope: ~40 well-known startups. Ask about valuation, sector, funding stage, HQ, headcount, or “similar to” — not general web questions.

Behind the demo

This mirrors a search engine I designed and shipped in production.

What you just used is a scaled-down version. The real one I owned end-to-end ran over a large production dataset — a query DSL that unified the stack, a compiler to optimized SQL, a vector layer for similarity search, and an LLM layer that turns a prompt into a structured query and refines it when ambiguous.