Retrieval-Augmented Generation (RAG) is a method that combines searching a private knowledge base with generating responses grounded in the found information, addressing a key limitation of Large Language Models (LLMs) which are generalists with no knowledge of a company's internal documents and can produce fabricated answers, a phenomenon known as hallucination. RAG works in a two-phase process. Phase 1, building the index, is done once: company documents are broken into segments, each converted into a numerical representation called an embedding vector that captures meaning. Phase 2, answering a question, happens every time: a user's prompt is converted into the same format and matched against the index by meaning similarity, with the closest segments retrieved and handed to the AI model to generate a grounded answer with citations. In practice, a product manager with thousands of pages of research documents, including interview transcripts, support tickets, and app reviews, can use a RAG system to obtain answers in seconds, such as a query for “reimbursement complaints,” instead of manual searches requiring hours of synthesis work.