Desenvolvimento de um Assistente Chatbot Inteligente para Instalações Elétricas baseado em Modelo de Linguagem Grande (LLM)

Abstract

This work deals with the development of an intelligent chatbot assistant for electrical installations, using a Large Language Model (LLM) combined with the Recursive Abstractive Processing for Tree-Organized Retrieval (RAPTOR) indexing technique, in order to provide accurate and efficient answers in accordance with the standards of the Brazilian Association of Technical Standards (ABNT) and Brazilian Regulatory Standards (NBR). The main aim is to make it easier for electrical professionals to interpret and apply technical standards, from drawing up projects to maintaining and inspecting installations. The methodology adopted involves several stages, including initial analysis and planning, definition of objectives, data collection and indexing, transformation of texts into embeddings, storage of embeddings in a vector database, and implementation of the text retrieval and generation pipeline. The results indicate a significant improvement in the efficiency and accuracy of the responses generated by the chatbot, contributing to the dissemination of technical knowledge and compliance with security regulations. Final considerations highlight the potential for future expansion into other regulatory domains.