Training & learning — RAG explained: a chatbot that uses your documents

Learning path

1. 1

   Search before generating

   RAG retrieves relevant passages before asking a model to answer. This adds document context without retraining the model for every update. Search can combine keywords and semantic similarity to match the meaning of a question.

2. 2

   Example: an internal procedure

   An employee asks how to request equipment. The assistant finds the approved procedure, explains the steps and links the source. If nothing relevant is found, it should say so. Citations help verification but do not guarantee correctness.

3. 3

   Respect document permissions

   Filter documents by user permissions before retrieval reaches the model. Remove obsolete versions and test questions with known answers. Treat instructions inside retrieved files as untrusted data, not permission to perform actions.