RETRIEVAL-AUGMENTED GENERATION (RAG)

LEARNING GOALS In this chapter, you will learn one of the most widely used techniques in practical AI systems. It is called Retrieval-Augmented Generation (RAG). You will learn what embeddings are, what a vector database is, why chunking matters, how to judge retrieval quality, and how all this helps reduce made-up answers. THE OPEN-BOOK EXAM PICTURE Imagine two students taking an exam about a company's vacation policy. The first student must answer entirely from memory, and she half remembers the policy, so she guesses and sounds confident. The second student is allowed an open-book exam. He quickly finds the page about vacation in the policy handbook, reads it, and answers based on what the page actually says. The second student will almost always be more accurate. RAG gives an AI model the open-book exam.