COMPARATIVE ANALYSIS OF AGENTIC RAG APPROACHES IN THE CONTEXT OF BUILDING A CONVERSATIONAL ASSISTANT

Abstract

As part of this paper, we present research and the implementation of two versions of a conversational assistant. The focus was on exploring Retrieval-Augmented Generation (RAG), which uses external knowledge in addition to built-in knowledge to generate responses and interact with users. The implementation started with a simple version that combines retrieval and generation. Afterwards, a more advanced approach was developed using agents and specialized tools for query processing. The goal of the paper is to provide a detailed description of the system development process, compare the architectures in terms of complexity, scalability, performance, and answer relevance. Initially, system evaluation was performed manually, while later performance analysis was automated using the Ragas framework. Finally, the advantages and disadvantages of both approaches are highlighted, along with scenarios where each would be appropriate. The paper concludes with suggestions for improvements and directions for future research.