Research

HOTSPOT (Hybrid Oceanic Tracking via Satellite Proxy and Optimized Time-series)

Studying how remote sensing and time-series modeling can be used to infer nutrient-driven ecological changes in marine environments.

Designed ecological constraint filters and local time-series interpolation to reconstruct missing data and enforce physically consistent predictions.

Focused on reducing false positives and improving reliability so environmental forecasts remain usable in real-world monitoring settings.

Earned the Stockholm Regional Water Prize, placed 3rd at regionals, and advanced to the state science fair.

AI Interpretability Research: Mechanistic Interpretability

Studying how neural networks represent concepts internally using modern interpretability methods.

Analyzed feature discovery, probing, activation patching, and weight-based analysis to understand how models encode and manipulate information.

Focused on representation geometry, sparse feature directions, and toy models that reveal mechanisms like superposition and grokking.

Synthesized 30+ papers on modern mechanistic interpretability as part of the NCSSM Research in Computational Science program.