Skip to content
Yarrangobilly

Primer

What is RAG in AI?

Retrieval-augmented generation grounds an answer in documents you control.

A language model is trained on a wide public slice of the internet. It does not automatically know your contracts, your SOP, or last week's inventory report. Retrieval-augmented generation, or RAG, means the system searches a collection you chose, pulls the passages that look relevant, and writes an answer with those passages in view.

That is why teams reach for it in document-heavy work: policies, claims files, trade confirmations, routing guides. It is not magic, and it is not the same as training a new model. If the collection is messy, outdated, or missing the record that matters, the answer will be too. Access control has to follow the documents. A search that can see every file is a leak with a nicer interface.

RAG also is not always the right tool. Some jobs need a calculation, a workflow, or a write back to a system of record. In those cases retrieval is a step, not the product. Start with the question a person already asks, the folder they already trust, and a way to see which passage the answer came from.