ML Engineer @ eBay · fraud and anomaly detection · applied ML at scale
I build applied ML systems from messy, real-world data, and I contribute to the open-source anomaly-detection stack in Python.
Now · anomaly detection, fraud ML, and cleaner ways to ship and evaluate models.
- aeon: authored the MADRID multi-length discord detector, shipped in v1.6.0 (listed maintainer); DAMP detector open
- PyOD: save/load/clone round-trip test coverage across 23 detectors (merged); Deep SAD and Extended Isolation Forest detectors open
- category_encoders: fixed
GrayEncoder.inverse_transform(merged) · scikit-dimension: uncentered lPCA option (merged) - river: Robust Random Cut Forest and a
PredictiveAnomalyDetectionfix (open) · scikit-learn: exact IsolationForest path length (open)
- graphspot: my own library for inductive graph anomaly detection with a torch-free core (tree and spectral detectors, Fraudar, OddBall, leakage-safe temporal evaluation), on PyPI
- social-unrest-forecasting: SUB-Forecast, a leakage-controlled benchmark for forecasting unrest in 994 regions across 55 countries; news text is informative but nearly subsumed by the event record
- GeoSocialX: geospatial toolkit for geotagged social data (CSV · GeoJSON · Bluesky), with mapping built in