Techniques for Improving the Effectiveness of RAG Systems
, Senior Solutions Architect, NVIDIA
Retrieval-augmented generation (RAG) systems show great promise for business applications, but a naive approach simply pairing a vector database retriever with a general-purpose large language model (LLM) rarely leads to high-quality results. Learn techniques that can take your RAG system from an interesting proof-of-concept to a serious asset. We'll cover hybrid retrievers, using multiple smaller fine-tuned expert models instead of a single large general-purpose model, and a cognitive framework you can use to make decisions about the best components to add to your RAG application. We'll show you how to evaluate RAG performance with each iterative design change, using both human-as-a-judge and LLM-as-a-judge evaluation frameworks to measure the impact of these techniques vs. a naive RAG baseline. With the lessons learned in this workshop, you’ll be able to build applications that deliver on the expectations of what serious LLM-based RAG applications can do. Prerequisite(s):