Hybrid Search

Hybrid search is a retrieval technique that combines keyword-based search — typically BM25 over an inverted index — with vector-based semantic search, and fuses the two rankings into a single result list. Keyword search excels on exact terms, rare words, product codes, and acronyms that embeddings tend to smear together; vector search excels on paraphrased queries and synonyms where the user's words do not appear verbatim in the document. The two signals are combined via weighted score fusion or reciprocal rank fusion, where each document's final score depends on its rank in both lists. Hybrid search is often the practical default for production RAG because real user queries come in both styles — some look like keyword searches, some look like natural-language questions — and a single retriever rarely handles both well.