GridGenius is built on a modular, scalable architecture. The backend uses FastAPI to serve RESTful APIs for model inference and RAG-based explainability, with data retrieval from a ChromaDB vector database. The frontend, hosted on Vercel, features an interactive dashboard built with Tailwind CSS and Vanilla JS, displaying forecasts, visualizations, and a chatbot interface powered by Groq Llama 3 with HuggingFace embeddings.
The system employs multiple ML models, including Random Forest, XGBoost, and a custom Transformer Regressor for demand forecasting. Feature engineering includes outlier removal, scaling, and derived features like Demand-Generation Gap. Extensive EDA revealed key insights, such as demand increases during hotter months and reduced demand on holidays, informing the model's accuracy and usability.