Atlas Semantic
Edge-deployed hybrid search across 80M enterprise documents
Atlas Semantic is a hybrid retrieval engine that indexes 80 million internal documents and serves sub-200ms queries from the edge — finally giving a global workforce a search box that actually finds things.
The challenge
The previous engine was an over-tuned BM25 index that hadn't been touched in five years. Knowledge workers spent an hour a day failing to find policies, decks and contracts they knew existed.
Our approach
We layered dense embeddings over the existing BM25 corpus, added cross-encoder re-ranking on the top 50 candidates, and pushed the serving layer onto Cloudflare Workers with regional vector replicas to hit sub-200ms anywhere in the world.
The outcome
Findability scores doubled, time-to-first-result fell 71%, and the search bar became the #1 entry point to the company knowledge base for the first time ever.
What we built
- Hybrid BM25 + dense embedding retrieval at 80M scale
- Cross-encoder re-ranking on the top-50 candidates
- Edge serving with regional vector replicas
- Permission-aware filtering at query time
CognitoFlow Engine
Autonomous supply-chain reasoning at Fortune 500 scale