Elasticsearch Vector Search

Elasticsearch provides dense vector search capabilities alongside its traditional full-text search, enabling hybrid retrieval for AI applications. Elastic Cloud offers 14-day trial; pricing based on deployment size. Self-hosted free with Elastic license. Vector search uses dense_vector field type with HNSW indexing for approximate k-NN search. Hybrid search combines BM25 text scoring with vector similarity through reciprocal rank fusion, powerful for text-to-SQL where keyword matches (exact column names) and semantic similarity both matter. ELSER (Elastic Learned Sparse EncodeR) provides trained sparse retrieval without embedding models. Features include filtering during vector search, quantization for storage efficiency, and byte vectors for reduced memory. For text-to-SQL: leverage existing Elasticsearch infrastructure for schema and query example storage with hybrid retrieval. Page should cover: dense_vector configuration, k-NN query syntax, hybrid search with RRF, ELSER for sparse retrieval, comparison with dedicated vector databases, and implementation patterns for RAG applications.