AI-Powered Electricity Price Forecasting & Purchase Recommendation System

End-to-end system built for Sofrecom Tunisia (Orange Group) to forecast French electricity calendar prices and recommend optimal purchase timing. It combines a deep-learning forecasting model (CNN-LSTM-Attention), a news-analysis AI agent, and a purchase-recommendation agent, all served through a FastAPI + React dashboard with an automated daily data pipeline. The model reached a ~4-5 €/MWh error on Cal 2027 and outperformed standard baselines (SARIMAX, Prophet, LightGBM, XGBoost, LSTM). Adding news analysis cut purchase-recommendation error by about 40%.