What is RAG?
Retrieval-augmented generation (RAG) fetches relevant passages from your own documents at question time and gives them to a language model, so answers draw on current, specific sources.
In more detail
A model on its own knows only what it was trained on. RAG adds a lookup step. Documents are split into chunks, turned into embeddings and stored in an index. When someone asks a question, the system finds the closest chunks, puts them in the prompt and asks the model to answer from them, ideally with citations. That keeps answers current without retraining and lets you restrict who sees which documents. Quality depends mostly on retrieval, not the model. Bad chunking, stale files or missing permissions produce confident wrong answers, so we test with real questions and expected sources before anyone relies on it.
Example
A consultancy indexes thousands of past proposals and statements of work. A new bid manager asks, 'What did we charge for a public-sector data migration?' and gets three relevant clauses with links to the originals.