Projects
Competitions and projects, each written up with how it was validated, how good the model is, and where it is weak.
Hobby project, Reading app, 2026
Reading Tracker: two buttons and a dashboard for my reading habit
A hobby app for my own reading. Two buttons log each reading session, and a public dashboard turns the sessions into streaks, pace and an estimate of when I will finish my current book.
1 day
To the first version
Competition, DrivenData, 2026
Two views of a brain scan: a calibrated 3D CNN and tree stack for DaT-SPECT
A stacked 3D CNN ensemble and gradient-boosted tree model that separates abnormal from normal DaT-SPECT scans with out-of-fold AUROC 0.9455. The CNN ensemble's probabilities are calibrated to an ECE of 0.0187, and the baseline's calibration carried over to held-out acquisition families at a log-loss cost of +0.0006.
0.9455
OOF AUROC, final model
Project Placeholder
Project title
One line that leads with the model and its validated result.
Competition Placeholder
Project title
One line that leads with the model and its validated result.