RAG in Plain English
Retrieve-then-generate; what it is, and when it beats fine-tuning.
A model only knows what it learned during training. It does not know your documents, your latest numbers, or anything private. Ask it about those and it will either guess or make something up. RAG is the fix.
How it works
RAG stands for retrieval-augmented generation, which is a fancy name for "look it up first, then answer." It adds a step before the model responds:
- Take your question.
- Search your own material for the most relevant pieces, usually by meaning, using embeddings.
- Hand those pieces to the model along with the question.
- The model answers using what you gave it.
The model is still doing the writing. It is just writing from your facts instead of its memory.
RAG versus fine-tuning
People mix these up. Fine-tuning changes the model itself to bake in a style or skill; it is slow, costly, and bad at fresh facts. RAG leaves the model alone and feeds it the right context at question time. For "answer from my documents," RAG is almost always the cheaper, more current, more reliable choice.
Why it matters
RAG is how most useful AI apps stay correct: support bots, document search, anything that has to be right about specific, changing information. It is the difference between a model that sounds confident and one that is actually right about your stuff. (For the real-world wiring - vector store, service API, the whole layer - see Service-Side Infra.)