RAG-VECTOR
Retrieval+LLM fact-grounded AI pattern.
Definition
RAG architecture: (1) Document chunking, (2) Embedding via OpenAI ada-002 / Cohere / BGE, (3) Vector DB storage (Pinecone, Weaviate, Qdrant, pgvector), (4) Query embedding, (5) Top-k similarity search, (6) Context injection in LLM prompt, (7) Answer generation. Hybrid Search (BM25 + vector) often superior. Use cases: support chatbots, knowledge base, enterprise search.
Origin
RAG formalised in Meta AI Research paper Retrieval-Augmented Generation (Lewis et al, May 2020) ; popularised by LangChain 2022 and GenAI adoption.
Example in context
An enterprise legal assistant indexes 50000 contracts in 500-token chunks in Pinecone ; query indemnity risk supplier contract returns top-5 chunks injected into GPT-4 prompt for sourced answer.
Related terms
- BM25 — hybrid search component.