Summary

This chapter taught us how to build an advanced RAG inference pipeline. We started by looking into the software architecture of the RAG system. Then, we zoomed in on the advanced RAG methods we used within the retrieval module, such as query expansion, self-querying, filtered vector search, and reranking. Afterward, we saw how to write a modular ContextRetriever class that glues all the advanced RAG components under a single interface, making searching for relevant documents a breeze. Ultimately, we looked into how to connect all the missing dots, such as the retrieval, the prompt augmentation, and the LLM call, under a single RAG function that will serve as our RAG inference pipeline.

As highlighted a few times in this chapter, we couldn’t test our fine-tuned LLM because we haven’t deployed it yet to AWS SageMaker as an inference endpoint. Thus, in the next chapter, we will learn how to deploy the LLM to AWS SageMaker, write an inference interface to call...