Retrieval-Augmented Generation
An AI architecture that grounds language-model outputs in retrieved source documents, constraining responses to information actually present in the knowledge base.
Retrieval-Augmented Generation (RAG) combines a document-retrieval system with a large language model. When a question is asked, the retrieval system surfaces the relevant passages from a curated knowledge base, and the language model is constrained to compose its answer from those passages — citing the source for each claim.
The contrast is with “free-generation” systems where the language model is asked the question directly with no retrieval grounding. Free generation produces fluent answers but cannot guarantee fidelity to a specific document corpus — which is why it produces hallucinations in regulated workflows.
RAG is the architectural pattern that makes AI deployable in regulated workflows. Every answer carries a citation back to the underlying source, which is the standard a compliance officer can verify. It is also why “AI hallucination” is a solved problem for any vendor who has actually built on a RAG pattern — and an open problem for vendors who haven’t.