Published:
A Julia reimplementation of CE-QUAL-W2, USACE’s Fortran hydrodynamic and water-quality reservoir model — rebuilt for multi-core parallelism and gradient-based calibration, validated line-by-line against real reservoir data and the original ~45,000-line source.
Published:
Digital twin models that predict dangerous harmful algal blooms before they happen, combining watershed simulation, hydrodynamic modeling, and machine learning. Funded by U.S. Army Corps of Engineers.
Published:
A probabilistic catastrophe model that simulates 10,000 years of hurricanes against a synthetic $39.01B coastal property portfolio to price insured losses.
Published:
Universal Dynamic Equations (UDEs) that uncover hidden drivers of ecosystem collapse from observable data alone—delivering superior early-warning capability for regime shifts with less data than conventional methods.
Published:
Scalable Explainable AI framework achieving 84% label-transfer accuracy across 158 ImageNet categories, reducing annotation effort by 85%. Transforms opaque neural network predictions into human-readable part-based explanations.
Published:
Multi-stage LLM pipeline achieving 91% accuracy that converts plain-English questions into optimized SQL queries, cutting query fulfillment from days to under 3 seconds. Built during AI/ML internship at Seagate Technology.