Experience

• Reduced missed defects by 67% (held-out coverage 95.1% to 98.4%) at a fixed false-alarm budget, or false alarms 5.4× at matched recall, via bilevel Bayesian optimization of wafer-inspection recipes
• Discovered an image feature separating defects from nuisance at AUC 0.956 vs. the engine’s best 0.693 (200/200 bootstrap wins); showed Bayesian search recovers the production recipe in a median of 13 evaluations vs. 103
• Caught a coverage metric reporting a false 100% (85% of on-defect detections were on the wrong die), fixed it, and shipped 1,702 automated tests and an offline React + FastAPI dashboard

Translating the Office of Faculty Affairs Workflow at New York University with automation and AI integration

Improved the event detection precision from 68% to 91% and recall from 76% to 95% by developing ML models from scratch.

Conducted a comprehensive observability study to improve monitoring, performance, and reliability across enterprise applications.