Multi-agente IA conversacional para soporte al usuario en una empresa de Telecomunicaciones.

Abstract

Abstract: This work proposes the development of a multi-agent conversational system, aimed at optimizing the management of queries in an organizational environment. Based on an analysis of the queries made by users, it was identified that the greatest operational burden corresponded to repetitive and low-complexity questions, which allowed for the design of specialized agents aligned with the real needs of the department. The developed architecture incorporates an orchestrator agent responsible for identifying the user's intent and redirecting requests to the corresponding specialized agent. In addition, it integrates internal APIs and microservices to ensure consistency in its responses and system continuity. The use of natural language processing techniques allowed for the correct interpretation of user queries, even in the face of spelling errors, maintaining precise and natural responses. The validation of the multi-agent system was carried out through controlled functional tests and a usability evaluation based on the ISO 9241-11:2018 standard. The results show very good accuracy in query redirection, adequate response times, and adequate levels of effectiveness, efficiency, and user satisfaction, confirming that the multi-agent architecture constitutes a robust and scalable solution compared to approaches based on a single conversational agent.