Desenvolvimento de uma solução tecnológica utilizando APIs de inteligência artificial generativa para otimização de processos institucionais do IFPI

Abstract

Educational institutions maintain a large volume of normative and administrative documents whose access is hindered by dispersion, frequent updates, and complex language. This work presents an intelligent conversational solution for consulting information from the Federal Institute of Education, Science and Technology of Piauí, integrating generative artificial intelligence, retrieval-augmented generation, multi-agent systems, and a continuous knowledge base update mechanism inspired by belief revision theory. The architecture employs specialized agents for information retrieval, document processing, and knowledge consistency maintenance. The evaluation combined automated technical tests and user participation. Across 51 institutional questions, the system achieved a 100\% execution rate, 78\% grounding accuracy in institutional sources, an average semantic similarity of 0.78, a semantic accuracy of 63\%, and an average overlap of 0.54 between retrieved and expected sources. The belief revision mechanism demonstrated consistent behavior in 65 verification scenarios involving synthetic and real institutional documents. In the evaluation with 34 participants, the solution achieved an average score of 70.8 on the Chatbot Usability Questionnaire and an overall rating of 8.1 on a scale from 0 to 10. The results indicate that the approach is viable for supporting access to institutional information, continuous knowledge updating, response reliability, and user experience.