Large language models like GPT are brilliant, but have a flaw: by themselves, they can't answer questions about facts when they weren't trained on suitable data, and may "hallucinate" an answer. This is a problem for answering questions that require recent knowledge or domain knowledge. The solution is to use "retrieval augmented generation" by pairing the LLM with a data store. In this code-along, you'll learn how to perform retrieval augmented generation in a case study on movie data. You'll process movie data for storage in the Pinecone vector database, then use it to inform answers to questions to GPT. You'll also learn how to structure prompts in a way to assist passing knowledge to and from the LLM.