Predisolia: Time Series Forecasting for Solar Energy
Development of an advanced time-series forecasting system for photovoltaic production. The platform predicts solar energy output based on weather patterns and hardware telemetry, enabling better grid optimization.
MLOps, PyTorch & Cloud Architecture
- Time Series Modeling: Designed and trained advanced deep learning models using PyTorch to capture complex seasonalities and trends in solar production data.
- Automated Retraining Pipelines: Engineered continuous training (CT) workflows that automatically ingest new telemetry, evaluate data drift, and trigger model retraining to maintain high accuracy over time.
- Serverless Inference: Deployed the inference engine using Azure Functions, creating a highly scalable, event-driven architecture that delivers near real-time predictions to the client.
