Distant Stars, Real-Time Science

Summary

Data from several space-based exoplanet missions (Kepler, K2, TESS) has enabled the discovery of thousands of worlds, but much identification is still manual. With modern AI/ML, we automatically analyze large mission datasets to accurately identify exoplanets.

Our Mission

Mission. Build an AI/ML pipeline that learns from NASA’s open exoplanet datasets (Kepler, K2, TESS) to automatically identify exoplanets and provide a clean web interface for uploads, results and metrics. Why now. Transit photometry missions produced millions of light-curve events. Much of the vetting is still manual. With modern ML and careful preprocessing, we can accelerate and standardize discovery.

How it works.

Data: Ingest, clean and normalize features (e.g., orbital period, transit duration, planetary radius). Model: Gradient boosting baseline with strong cross-validation; confusion matrix and per-class metrics for transparency. API: FastAPI endpoints for CSV/JSON predictions, job status, artifacts and model info. UI: Static frontend to upload data, view results, and explore metrics—ready for Render/Ferozo. Impact. Faster triage of candidates, consistent labels across missions, and a foundation for human-in-the-loop review—helping researchers surface new exoplanets hidden in Kepler/K2/TESS data.

Background

Using the transit method, a planet crossing its star slightly dims the observed brightness. Kepler and K2 pioneered decade-scale transit surveys; TESS has continued all-sky monitoring since 2018. Public datasets list confirmed planets, candidates, and false positives with variables such as orbital period, transit duration, and planetary radius.

Objectives

• Train an AI/ML model on NASA’s open datasets to classify new observations as confirmed, candidate, or false positive.
• Provide a web interface for CSV uploads and JSON inputs, with instant results.
• Expose accuracy metrics, confusion matrix, and versioned artifacts for transparency.

Potential Considerations

• Support both researchers and newcomers with clear UX.
• Enable incremental training and hyperparameter tuning from the UI.
• Show live performance statistics and model governance info.