Term · Apps & software
RAG (retrieval-augmented generation)
RAG — retrieval-augmented generation — is a technique for building AI assistants that answer from specific sources, such as a company's documents, policies and product data. Instead of relying only on what a language model learned in training, the system first retrieves relevant passages and then asks the model to answer using them.
[ 01 ]
Explained
How it works
- Preparation: documents are split into passages and indexed — often as numerical representations ("embeddings") that capture meaning, sometimes combined with keyword search.
- Retrieval: for each question, the system finds the passages most likely to contain the answer.
- Generation: the language model receives the question and the retrieved passages, with instructions to answer only from them and cite sources.
- Guardrails: if passages do not answer the question, the assistant says so and offers a person.
Why businesses use it
- Answers reflect your current information, not general internet knowledge.
- Sources can be cited, so answers can be checked.
- Knowledge can be updated without retraining a model.
- It reduces — but does not eliminate — invented answers. See AI hallucinations and business risk.
Limits
- Quality depends on the knowledge base: outdated or contradictory documents produce poor answers. See preparing a knowledge base.
- Retrieval can miss relevant passages or bring irrelevant ones.
- Facts that change often — prices, stock — are better read from systems directly than from documents.
[ 02 ]
FAQ
Frequently asked questions
Is RAG the same as training a model on our data?
No. RAG supplies information at question time; training changes the model itself. RAG is usually simpler, cheaper and easier to keep current.
Does RAG work in Azerbaijani?
Yes, with models and retrieval that handle the language well; test on real questions.