Retrieval-augmented generation (or RAG) is an advanced technique for augmenting the capabilities of a generative AI model by integrating external knowledge bases that contain up-to-date, task- or domain-specific information. Through this approach, RAG improves the accuracy and relevance of generated content, reduces hallucinations, and enables more context-aware responses. In this chapter, I dive deep into the solution architecture and data flow of a general RAG solution. I’ll discuss the various options for implementing RAG in AWS, from the most managed (and abstracted) to the least managed (and least abstracted) solutions. I’ll describe and demonstrate using text and multimodal embedding models in Amazon Bedrock to create embeddings from text, images, video, and audio. I’ll then explore vector-supported data systems used for storing vectors, including cost-optimized options like S3 Vectors and more performant options like Aurora PostgreSQL with the PgVector extension and Amazon OpenSearch with the Vector plugin.