Lesson 1 of 7
What is RAG
6 min read
Ask a model "what's our API rate limit?" and it answers instantly, confidently — and makes the number up. It never saw your docs. So how do you get an answer from what's actually true?
Retrieve, then generate
RAG — Retrieval-Augmented Generation — is two steps. First retrieve: search your own sources for the passages that answer the question. Then generate: hand those passages to the model and ask it to answer from them. The model stops guessing from memory and starts reading from what you gave it.
RAG is retrieve-then-generate: find the right passages, then answer from them — not from the model's memory.
Why grounding beats memory
A model's built-in knowledge is fuzzy, out of date, and can't cite anything. Grounding — answering from sources you put in front of it — makes the reply accurate, current, and traceable: it can point to the exact passage it used. Change the source and the answer changes with it, with no retraining.
Grounding turns a confident guess into a checkable answer that points back to its source.
RAG doesn't teach the model new facts permanently — it hands them over at question time, for this one answer. Update the document and the next answer is current instantly.
The shape of it
- —RAG = retrieve relevant passages, then generate an answer from them.
- —Grounding makes answers accurate, current, and traceable to a source.
- —Change the source, change the answer — no retraining.
Your support bot keeps citing an old refund window. The policy doc was updated yesterday. With RAG, what fixes it?
Continue in the app
Take the whole RAG & Search course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.