Is RAG the same as fine-tuning?

No. Fine-tuning adjusts the model's weights and suits changing behaviour or style; RAG leaves the model untouched and feeds it facts at question time. For answering from your documents, RAG is cheaper, updatable in minutes when documents change, and able to cite sources.

What are the most common RAG failures?

Bad chunking that splits documents mid-thought, stale indexes built once and never refreshed, and prompt drift that quietly lets the model use general knowledge again. Each is cheap to prevent and expensive to discover in production.

// ai

RAG (Retrieval-Augmented Generation)

Retrieval-augmented generation (RAG) is a technique where the system fetches the relevant pages from your documents at question time, hands them to the language model along with the question, and has it answer from that material instead of relying on training memory. It is the standard pattern behind every "chat with your documents" product.

RAG matters because a language model's knowledge is frozen at training time and mixed with everything else on the internet; RAG narrows the job to answering from material you control. It is the most deployed AI pattern in business: employee knowledge bases, customer support from real manuals, contract lookup, technical documentation search. A well-built system answers only from your documents, cites the page it found, and says when the material says nothing — three properties a raw chatbot does not have.

Concretely: your documents are split into chunks, each chunk is converted into a numerical fingerprint (an embedding) stored in a vector database, and a question retrieves the closest chunks. At business scale the infrastructure is nearly free — vector storage and embeddings for tens of thousands of documents run in the tens of euros per month. The real costs are document cleanup, the ingestion pipeline and evaluation: a fixed set of fifty questions with known answers, rerun after every change.

Frequently asked questions

Is RAG the same as fine-tuning?: No. Fine-tuning adjusts the model's weights and suits changing behaviour or style; RAG leaves the model untouched and feeds it facts at question time. For answering from your documents, RAG is cheaper, updatable in minutes when documents change, and able to cite sources.
What are the most common RAG failures?: Bad chunking that splits documents mid-thought, stale indexes built once and never refreshed, and prompt drift that quietly lets the model use general knowledge again. Each is cheap to prevent and expensive to discover in production.

// go deeper

Full treatment in the guide: RAG, without the jargon: letting AI answer from your own documents

RAG (Retrieval-Augmented Generation)

Frequently asked questions

Related terms