Elasticsearch for hybrid search (text and vector)

Elasticsearch unifies BM25 keyword matching and kNN vector search into a single, high-performance query pipeline using Reciprocal Rank Fusion (RRF).

Modern search requires both precision and intent. Elasticsearch delivers this by running traditional BM25 text retrieval alongside dense vector k-nearest neighbor (kNN) searches in parallel. Instead of managing separate databases, you ingest text and let Elastic generate or store embeddings natively. The platform then merges these distinct scoring systems using Reciprocal Rank Fusion (RRF) (or linear boosting), producing a single, highly accurate ranked list that satisfies both exact keyword matches and conceptual, semantic queries.

https://www.elastic.co/guide/en/elasticsearch/reference/current/knn-search.html