GPT-Powered Chatbot for Dissemination and Simulation of Renewable Energy Communities

In this work, we present an architecture that seamlessly integrates a large language model (LLM) with MATLAB-based simulation code to support the design and analysis of Renewable Energy Communities (RECs). Users interact via a Telegram chatbot powered by a retrieval-augmented GPT-4.1-nano model, which translates natural-language requests into JSON inputs. These JSON payloads are consumed by a MATLAB 2024b-implemented REC simulator, which leverages PVGIS irradiation data, average domestic consumption profiles from the Italian regulatory authority, and 2024 day-ahead market prices to compute both individual and collective benefits under varied community configurations and load-shifting scenarios. Simulation outputs return as JSON, are parsed by the LLM, and rendered as human-readable summaries. This tight coupling combines the LLM's conversational flexibility and contextual understanding with conventional codes's deterministic numerical rigor, eliminating 'black-box' uncertainty in complex calculations while preserving natural-language usability. Case studies demonstrate robust performance, error resilience, and markedly improved JSON fidelity via prompt engineering.