Retrieval-augmented generation
RAG · A method in which a language model first finds the relevant passages in an organisation's own documents and then bases its answer on that information.
- 01
Why it matters
Retrieval-augmented generation lets the model answer with current, organisation-specific information without retraining, makes it possible to show the source of the answer and reduces the risk of made-up answers. Answer quality depends on how current the documents are and on finding the right passage. Users should only get answers from documents they are authorised to see.
- 02
Example
A shop's support assistant answers return questions from the current returns policy document and shows a link to the document under the answer. When the policy changes, only the document is updated.
- 03
Common mistake
Loading old and contradictory documents into the system without cleaning them up. The assistant may confidently tell a customer about a rule that is no longer in force.
- 04
Related terms
Related services and guides
Let's talk about your project.
Tell us what you need; we will define the scope together.