Real-time control of tokamaks — foundation models, computer vision, and signal processing for plasma state estimation and instability prediction. Advised by Egemen Kolemen at Princeton MAE / PPPL.
TokEye — fast signal extraction for fluctuating tokamak time series.
Current focus
Building foundation models for fusion: large neural networks trained on diverse tokamak data that can be specialized to downstream control and diagnostic tasks. I'm also working on real-time emission-front control and self-supervised identification of coherent modes — both run at DIII-D.
Selected experimental work
Led a real-time ML-enabled emission front control experiment at DIII-D (APS DPP 2024).
Built TokEye, a fast signal-extraction pipeline for fluctuating tokamak time series (arXiv Feb 2026).
Broader interests
Foundation models for scientific time-series data
Vision-based state estimation and active learning on streaming data
Uncertainty quantification when ML sits inside safety-critical control loops
Regulatory and compliance considerations for AI in fusion