Retrieval-augmented generation (RAG) fetches relevant records and places them in the model's prompt, so the answer draws on your data, not only on its training.
Category
AI
Also known as
RAG; retrieval augmented generation
It is the standard way to make a model answer from a product catalog, a help center, or a customer profile without retraining it. Quality depends heavily on the retrieval step: if the wrong records are fetched, the answer is fluently wrong.
In practice
Measure retrieval on its own. A model cannot ground an answer in a record that never reached the prompt, and many "the AI got it wrong" cases trace back to what was fetched.
Something out of date?