Having established how to retrieve relevant information in the previous chapters, we now focus on how that information is utilized. This chapter examines the 'Generation' part of the Retrieve-Augmented Generation process, specifically how a Large Language Model (LLM) uses the retrieved context to formulate a final answer.

We will cover the following key areas:

• The role and function of the LLM component within the RAG system.
• Techniques for constructing prompts that effectively incorporate both the original query and the retrieved context.
• Methods used to inject the found text passages into the LLM's input.
• Strategies for managing the context window limitations of LLMs when adding retrieved information.
• The process by which the LLM synthesizes the augmented prompt to produce a coherent response.
• Approaches for attributing the generated output back to the source documents, using associated metadata.

Upon completing this chapter, you will have a clear understanding of how retrieved data actively shapes and improves the LLM's final output in a RAG pipeline.