Hybrid search: meaning plus exact words

Your RAG app finds the paragraph about cancelling a subscription, but misses the one that mentions error code E-4021.

That's the classic gap in vector-only search. Embeddings match meaning. They're weak on things that have to match exactly: error codes, product names, IDs, rare terms, new acronyms. Keyword search (BM25) is the opposite. It nails exact terms and ignores paraphrase.

Hybrid search runs both and merges the two ranked lists. A common way to merge is reciprocal rank fusion. Each document scores by its rank in each list, and you sort by the total. You don't have to tune score scales, which is why people like it.

It helps most on technical docs, support tickets and legal text, and anywhere users search by code, SKU or ID.

A quick check: take 20 real queries that failed and see whether the right chunk was found by keywords but missed by vectors. If it was, hybrid will probably pay off fast. Many vector databases and search engines support both modes now, so it's often a config change, not a rebuild.

Have you tried hybrid on your own data? What changed?