AI-ORBIT Solar Monitoring
DeployedAgentic ML system for solar power plant anomaly detection
Fullstack - ML pipeline, dashboard, system integration · Group project (3 members) - AI course final
- Python
- XGBoost
- LSTM
- Docker
- Railway
- Streamlit
The Problem
Solar power plants need to catch performance anomalies early, but raw sensor data is noisy and hard to interpret manually.
What I Built
An agentic system combining 5 ML models with a 3-layer agent architecture, a real-time dashboard, and automated Telegram alerts, containerized with Docker.
Hard Problem I Solved
The models showed unrealistically high accuracy at first because of data leakage - the scaler was fit on the full dataset before the train/test split. I fixed the pipeline so scaling happens after the split, which gave honest, trustworthy performance.
What I Learned
How to integrate multiple ML models into one working system, and that building AI is mostly about the pipeline around the model - not just training it.
- ML models
- 5ML models
- agent layers
- 3agent layers
- alerts
- Livealerts