Two-Database Architecture: Vector DB + Graph DB

The recommended production architecture uses PostgreSQL with pgvector for vector semantic search and Neo4j for knowledge graph storage and traversal. PostgreSQL handles vector storage with HNSW indexes, full-text search, document storage, and hybrid search with Reciprocal Rank Fusion. Neo4j handles entity and relationship storage, community detection via GDS library, multi-hop graph traversal, and Cypher-based querying. The hybrid retrieval pipeline encodes queries as vectors, runs parallel vector search and graph traversal, fuses results using RRF with k=50, and passes context to the LLM.