FIFoundationIndex
IndexMapAbout
ConceptLLMCross-Modal

RAG & Hybrid Retrieval

Also known as: RAG · Retrieval-Augmented Generation · pgvector · BM25 · Reranking

Plain English

Giving an AI model a live search engine over your own documents, databases, or code before it answers—so responses stay grounded in up-to-date, verifiable facts instead of relying only on frozen training weights.

Technical Overview

Production Retrieval-Augmented Generation combines lexical search (BM25 / Postgres tsvector) with dense vector embeddings (pgvector, Qdrant, Turbopuffer) fused via Reciprocal Rank Fusion (RRF) and cross-encoder reranking. Even with million-token context windows, hybrid RAG remains vital for latency, cost (paired with prefix caching), access control, and citation faithfulness verified by Evals & Benchmarks.

Chronological Entries (2)

#1Ecosystem
Curated

Anthropic published Contextual Retrieval, showing that combining BM25 lexical matching, dense embeddings, contextual chunk headers, and reranking reduces retrieval failure rates by up to 67% compared to naive vector-only RAG.