Enhancing Retrieval-Augmented Generation Systems at NetApp

This white paper discusses the development and refinement of a Retrieval-Augmented Generation (RAG) system named 'Doc' at NetApp. It outlines the holistic approach necessary to enhance answer relevancy and accuracy in RAG systems, focusing on three primary areas: prompting strategies, retrieval mechanisms, and documentation improvements. The paper emphasizes that the effectiveness of a RAG system is influenced not just by the language model or knowledge base size, but by the balance between user question formulation, information retrieval, and documentation structure. It presents detailed strategies for improving each component, supported by real-world examples and data-driven observations from the development of Doc. The insights shared aim to assist developers, researchers, and organizations in implementing or enhancing their own RAG solutions. The paper also addresses the importance of context-aware prompting and optimization of retrieval mechanisms, including the use of Azure Cognitive Search and its relevance scoring features.