I am a 14-year-old developer and I have finalized a python-based data pipeline designed to solve the extreme class imbalance (~1.3% positive class ratio) in radio astronomy telemetry (HTRU2 dataset validation). Standard LightGBM baselines optimize for global accuracy, causing catastrophic coverage failure for rare minority events (dropping true pulsars). Heuristic reweighting controls this but spikes False Positives, wasting expensive telescope pointing hours.
To solve this, my software builds a class-conditional non-conformity score framework using Mondrian Conformal Prediction. It splits live telemetry streams into mathematical certainty tiers, achieving a statistically verified 96.05% pulsar coverage while guaranteeing a strict error floor (1 - α = 0.05).
The engineering is complete, and the code repository is fully public. I am trying to submit this pipeline to the IRIS National Science Fair pool before the strict October 3rd deadline. However, my school cannot process or clear the mandatory ₹5,900 registration processing fee.
I am looking for structural code critique, technical mentorship, or guidance from the community on how to align this repository with a micro-grant or ecosystem backing to clear the submission desk.
If interested, here is my GitHub link - GitHub - Not-kh-lily-23/pulsar-conformal-triage · GitHub