Techniques & Methods
Hybrid Search in plain English.
Also known as: hybrid retrieval,sparse plus dense retrieval,keyword plus vector search
The one-sentence version
Combining keyword search and vector similarity search in one query so results match both exact terms and meaning.
Hybrid search runs two kinds of retrieval at once and merges the results. Keyword (sparse) search, typically BM25, is excellent at exact matches such as product codes, names, and rare terms. Vector (dense) search finds passages that mean the same thing even when the words differ. Each fails where the other succeeds: vector search can miss a specific error code, and keyword search cannot tell that "cancel my plan" and "end subscription" are the same request. Hybrid search scores both, combines them (reciprocal rank fusion is the common method), and often passes the merged list through a reranker for a final ordering. Most vector databases now support it natively: Weaviate and Qdrant ship sparse-vector support, and Pinecone offers sparse-dense indexes. For RAG systems over technical documentation, legal text, or anything with identifiers, hybrid search is usually a clear improvement over pure vector retrieval and has become the default recommendation.