Atlas Docs

A semantic search engine for developer docs, combining Elasticsearch with vector embeddings.

The challenge

The Atlas Foundation’s documentation had grown to 9,000 pages, and its keyword search rarely found what developers actually meant.

What I built

I built a hybrid search service in Python that blends Elasticsearch keyword matching with vector embeddings, plus a fast React search UI with instant previews.

The result

40k developers search the docs every month, the share of searches with no useful result fell from 32% to 6%, and support questions dropped by a third.