Retrieval-Augmented Generation (RAG) is an architecture in which a generative model receives information retrieved from external sources before producing an answer.
How it works
A typical flow includes document indexing, retrieval of relevant content and injection of that context into the model prompt. The goal is to use current or proprietary knowledge rather than relying only on model parameters.
Practical example
An internal assistant can search company procedures, policies and manuals and generate an answer grounded in the retrieved passages, citing relevant sources.
Why it matters
It is useful for knowledge management, customer service and decision support when information changes or should not be embedded in model training.
What to watch
RAG does not guarantee correctness: if retrieval returns wrong or incomplete documents, the answer can still be wrong. Retrieval quality must be evaluated.