GLOSSARY
Retrieval augmented generation (RAG)
Retrieval-augmented generation, or RAG, adds relevant retrieved information to a model request. The material may come from product documentation, approved knowledge collections or other sources. The answer depends on both retrieval and subsequent use of the material.
Apply product knowledge to a customer case
A salesperson wants to establish whether a proposed feature meets a customer's specific requirement. RAG could supply relevant document passages. If retrieval finds an outdated product version or the answer misses a qualification, a well-written response can lead to an incorrect commitment.
Validity, product scope and permissions therefore matter during information selection. Users benefit from seeing which passage supports a statement and where the available material provides no answer. Maintaining the knowledge collection becomes an operational responsibility, with an owner for obsolete or conflicting documents.
Evaluate retrieval and response separately
First test whether the necessary passages are found. Then check whether the answer preserves their conditions. Include conflicting versions and missing information in the test cases. A vector database can support retrieval. The decisive quality criterion remains the evidence supporting the particular claim; attaching a link does not by itself establish that support.