Retrieval-Augmented Generation (RAG)

An AI system architecture in which a large language model's answer is grounded in documents retrieved, at query time, from a searchable corpus, rather than relying solely on facts encoded in the model's weights during training. RAG lets a system stay current with fast-changing information without retraining, at the cost of making the retrieval corpus itself a live, consequential production data system.

A pattern in which the system searches a chosen set of documents, pulls the most relevant passages, and inserts them into the prompt so the model answers from that material rather than only from its training. It is the main reason the same model gives two organizations different answers, because each retrieves from its own documents (named and demonstrated by Lewis et al., NeurIPS 2020).

A technique in which a large language model's response is grounded in a specific, verified source document provided to it at the time of the query, rather than relying solely on patterns learned during training. RAG meaningfully reduces, but does not eliminate, hallucination risk, because the model can still misinterpret or inaccurately summarize the source it was given.

An agent architecture that searches a document index or knowledge base for content relevant to a query, inserts that content into the model's context, and generates an answer grounded in what was retrieved; a common source of scoping failures when the search index is not filtered by the requesting user's permissions.

An architecture that retrieves relevant, current passages from a connected source and feeds them into the prompt so the model answers from real supplied material rather than only its training memory. It is the dominant enterprise method for closing the context deficit and reducing hallucination.