Geração de memoriais de cálculo elétrico residencial via agentes LLM e RAG: uma abordagem baseada na NBR 5410

Abstract

This work presents the development and validation of an agent based on Large Language Models(LLMs) and Retrieval-Augmented Generation (RAG) to assist in the initial sizing of low-voltage residential electrical projects. The proposed methodology integrates natural language interaction and floor plan interpretation with deterministic calculation routines, ensuring compliance with the ABNT NBR 5410 standard and local utility guidelines. The system was structured to perform load forecasting, terminal circuit division, conductor and protection device sizing, and supply pattern definition, automatically generating an auditable calculation report. Validation was conducted through case studies based on didactic engineering exercises, demonstrating that the agent is capable of producing consistent, normatively compliant, and electrically coherent results, offering an effective tool for increasing productivity in the preliminary stage of electrical projects.