RAG

RAG, retrieval-augmented generation, is how you get a model to answer from your data instead of only what it learned in training. When a question comes in, the system first searches your documents for the relevant bits, then hands those to the model along with the question, and the model answers grounded in what it just read. It's the reason a chatbot can cite your policies or pull from last week's numbers.

The appeal is that you don't retrain anything. You point it at your documents, and as those documents change, the answers change with them. It also cites sources, so you can check where an answer came from, which matters when accuracy is on the line.

Use RAG when the model needs to know things it wasn't trained on: your internal docs, or anything that changes often. It's usually the first thing to try when you want an AI grounded in specific, current information, since it's cheaper and faster to stand up than training a model.

The catch is that RAG is only as good as what it retrieves. If the search pulls the wrong passages, the answer suffers, so the quality of your documents and the retrieval step matters a lot. When you need the model to adopt a specific behavior or format rather than specific facts, that's a job for fine-tuning, and plenty of systems use both.