Hybrid search is the retrieval stage of a modern RAG pipeline: run a semantic (vector) search and a keyword search in parallel, then merge their rankings. It fixes the blind spot each has on its own.
Why one method isn't enough
Vector search finds meaning but fumbles exact strings — error codes, SKUs, function names. Keyword search (BM25) nails exact strings but misses paraphrases. Real queries need both, so you run both and combine the results.
How Reciprocal Rank Fusion works
RRF is a simple, robust way to merge two ranked lists: each result scores 1 / (k + rank) in each list, and you sum the scores. A passage that ranks well in both rises to the top; one that ranks highly in only one still gets credit. No score normalization, no tuning weights — it just works, which is why it's the default.
Then rerank the top of the merged list
Fusion gets the right passages into the top 20; reranking for RAG reorders those so the single best passage is first, letting you pass fewer chunks to the model. Kit for AI runs hybrid search with RRF and reranking in-house on every query — Start free and see the results.