Performance Enhancement of Agentic Retrieval Augmented Generation Using Relevance Generative Answering

The aim of this research paper is to present a novel approach of using Relevance Generative Answering (RGA) in the trending field of Agentic Retrieval Augmented Generation (RAG). The paradigm shift in the RAG system by the introduction of Agentic RAG has opened a new research paradigm. The major issue of hallucination is overcome with the use of a traditional RAG system with some limitations like accuracy and relevance, lack of reasoning, the lost in the middle problem, etc. The Agentic RAG system attempts to address a few of these limitations. However, interpreting results based on the user's intent remains a significant area of research. This research aimed to understand user intent by introducing relevance detection block in the proposed architecture. Different performance metrics like precision, recall, F1 score, relevance, latency are used to validate the proposed approach. The results presented in this research reveal that the performance of the proposed system is much more relevant compared to agentic RAG system. For context and intent specific applications proposed framework suits well.