How do you build a simple but useful RAG?
Answer out loud or in your head first, then compare.
Answer
I separate preparing the documents from handling a question. First I fetch the content and permissions, split the documents and store them in the search engine. When a question comes in, I check the user, retrieve the passages they can access, and only then pass them to the model.
The answer should point to sources and be able to admit that material was missing. I need a way to update and delete documents. To start, I compare results with a simple search engine and a few hand-described tasks. I don't begin with multiple agents. A good first scope can cover one type of document and reading only, without performing actions. Access control is part of that scope, not an add-on at the end.
Why it matters: Design after learning RAG's parts, not guessing tool names.
Possible follow-ups: How will you check whether the problem lies in retrieval or in the model?
How did it go? Be honest — “Not yet” puts the question on your review list.
Start with
- G-P01 · What is RAG? — RAG and retrieval
- G-P02 · What are embeddings? — RAG and retrieval
Further reading
- Azure AI Search: Hybrid search overview(opens in a new tab)Docs; accessed 22.10.2026. Vectors plus full text, with rankings merged by RRF.
- Azure AI Search: Chunk large documents(opens in a new tab)Docs; accessed 22.10.2026. Choosing chunk boundaries and keeping context.