Desenvolvimento de um sistema conversacional multiagente com recuperação de conhecimento (RAG) e modelos de linguagem (LLMS) para suporte ao estudante na UTFPR-CP

Abstract

In the current scenario, the growing demand for quick and frequent access to institutional information at the Federal University of Technology - Paraná, Cornélio Procópio Campus, highlights the need to modernize student support. The dispersion of data across multiple systems and documents makes it difficult to obtain accurate and up-to-date information, overloading administrative departments and compromising the student experience. To address this challenge, the development and implementation of an intelligent virtual assistant is proposed, capable of centralizing and contextually processing the institution's public information. The solution is based on the integration of advanced Natural Language Processing (NLP) techniques, Large Language Models (LLMs), and the Retrieval-Augmented Generation (RAG) approach. The adopted multi-agent architecture leverages the LangChain library for agent orchestration and the Docling tool for processing and transforming institutional documents into semantically indexed vector representations. This system combines language generation capabilities with information retrieval mechanisms, enabling precise and relevant queries through optimized vector-based knowledge retrieval. With the implementation of this system, it is expected to significantly improve access to information, reduce the workload of administrative departments, and enhance the quality of academic support provided to students, aligning with the contemporary demands of digital transformation in the university environment.