Retrieval-Augmented Generation (RAG)

Retrieval-Augmented Generation, commonly called RAG, combines a generative model with an external retrieval system. Instead of relying only on information encoded in model parameters, a RAG pipeline searches a knowledge source and provides relevant material to the model before generation.^[1]

A typical pipeline converts documents and queries into representations, performs Vector Search or another retrieval method, selects relevant passages, and inserts them into the prompt. Embedding models are frequently used for semantic retrieval, although keyword and hybrid retrieval can also be effective.

RAG can improve freshness, traceability, and domain grounding, but it is not an automatic truth mechanism. Retrieval quality, document quality, prompt construction, and citation handling all affect the final answer.

See also Embedding, Vector Search, Large Language Model (LLM), and hallucination.