Also known as: RAG
In plain English
RAG is like an open-book exam. Before answering, the AI searches your documents for the relevant pages and uses them, instead of relying only on what it remembers.
In practice
RAG is the most common way to make AI answer questions about company knowledge, such as policies, product manuals and past tickets. It reduces hallucinations, keeps answers current without retraining, and can cite sources. Answer quality depends heavily on how documents are split and searched.
Under the hood
A RAG pipeline splits documents into chunks, embeds them into a vector index, retrieves the best matches for each query (often with hybrid search and re-ranking), and inserts them into the prompt. Evaluation covers both retrieval quality and the faithfulness of the generated answer.
Example
"The HR assistant uses RAG to answer leave questions from our policy documents."