// CASE STUDY : APPLIED ML : SPACE : 2026

ExoDetect

A model that flags an exoplanet but will not say why is not evidence an astronomer can act on.

ROLE: Product : ML
CONTEXT: ISRO Build-a-thon 2026, PS7
DATA: NASA Kepler : TESS
STATUS: Live demo

The problem

Speed is not the bottleneck. Trust is.

Machine learning can sift Kepler and TESS light curves far faster than manual review, and that part is close to solved. The problem is what happens next: an unexplained classification cannot be defended, published or followed up, so it does not actually reduce anyone's workload.

My thinking

Being written.

The solution

Being written.

Execution

Being written.

Impact

91% CLASSIFICATION ACCURACY WITH EVERY PREDICTION INTERPRETABLE

91%

classification accuracy

98.1%

ROC-AUC

What I learned

Being written.

Still being written - the sections marked “Being written” are next.