Sistema multi-agente basado en LLMs para la categorización y enrutamiento inteligente de tickets de soporte tecnológico universitario

Abstract

Abstract: This curricular integration work addresses the problem of the manual management of university technology support tickets, proposing the development of a multi-agent system based on Large Language Models (LLMs). The solution was implemented for the Universidad Técnica Particular de Loja by automating the lifecycle of its helpdesk incident tickets, from reception, classification and subclassification, prioritization by urgency level, and routing to Level 1 or 2 technicians; in addition to having the capability to automatically respond to tickets that the system believes it can resolve. Through a hexagonal architecture coordinated by an orchestrator agent and different specialized agents, using Retrieval-Augmented Generation (RAG) techniques, the system is able to interpret user intent and resolve frequent inquiries without human intervention. The results obtained demonstrate that the integration of autonomous agents optimizes response times and reduces repetitive tasks for Level 1 technicians, validating the use of LLMs to improve efficiency and user experience at the technology service desk.