Retrieval-augmented generation
Also: RAG, retrieval augmented generation
A pipeline that chunks and embeds a document collection, embeds each query into the same vector space, retrieves the nearest chunks and passes them with the question to an LLM. Answers can then use private or up-to-date facts and cite sources, which reduces hallucination.
Example
Asked "What is our refund window?", the system embeds the question, retrieves the 3 most similar chunks from the policy handbook, and the LLM answers "30 days" citing the chunk it came from, a fact it never saw in training.