About
I build machine learning systems for high-dimensional time-series — mostly clinical neural signals, sometimes industrial sensors. Over 15 years I’ve taken models from prototype to deployment: a patented rehabilitation algorithm that is licensed and clinically deployed, an NIH-funded real-time decoding system used in stroke-rehabilitation trials, and production ML at Generac Power Systems.
Day to day: deep learning in PyTorch (transformers, self-supervised learning, domain adaptation) on messy real-world data, with an emphasis on rigorous validation and reproducible pipelines.
Shipped
- NIH R01 clinical ML — real-time EEG decoding for a stroke-rehab exoskeleton, validated on a heterogeneous patient cohort (Frontiers in Neuroscience, 2016).
- Compact CNN-Transformer — 73% fewer parameters at parity accuracy for edge inference (Scientific Reports, 2024).
- Production industrial ML — anomaly detection and predictive maintenance.
- Open-source BCI framework — 100+ GitHub stars, used by 5+ external groups.