This chapter explains how the system generates answers using both internal documents and a large language model.
The system now performs retrieval-augmented generation.
It retrieves relevant documents for a user query.
Then combines them with the query to generate a response.
This method gives better results than directly asking the LLM.
The extra context from retrieved documents improves accuracy and relevance.
However, hallucinations can still happen, and the chapter introduces ways to reduce them.
Engineers are advised to focus on three key areas:
Ensuring document quality.
Structuring prompt inputs properly.
Evaluating the LLM output against expected answers.
The chapter emphasizes engineering responsibility in tuning RAG systems for high reliability.