ISRO Hackathon Dataset

The repository contains raw and processed data, training scripts, model artifacts, and a documented FastAPI deployment path for space-weather forecasting.

Space-weather signals change over time and are difficult to reason about from a single model or single forecast horizon. The project organizes feature engineering, training, and serving around that problem.

The repository evidence points to these observable boundaries and technologies:

• Raw and processed dataset layers
• Physics-informed feature engineering
• Random Forest, XGBoost, and PyTorch model paths
• Multi-horizon prediction targets
• FastAPI deployment package and LFS-tracked artifacts

• Training scripts cover XGBoost, Random Forest, and neural-network paths.
• The README documents 23 physics-informed features and 45-minute, 6-hour, and 12-hour targets.
• The public profile links to a Hugging Face Space, which responded successfully during the audit.

• Keeping large data and model artifacts organized with Git LFS.
• Turning physical signals into features that can support more than one forecast horizon.
• Separating documented benchmark claims from results independently reproduced in this portfolio.

• The repository contains raw data, processed data, model files, and training scripts.
• The public Hugging Face Space link from the profile responded with HTTP 200 during the audit.
• No additional performance number is presented here beyond what the repository documents.

The project demonstrates the full path from data organization to model serving while keeping the technical story grounded in observable artifacts.