Noise Geometry Determines Audit Efficiency
GitHub ↗A study of grokking and generalization that asks how the geometry of structured label noise decides how efficiently a limited audit budget can relabel data and recover clean performance.
Chemical Biology + Computer Science
I'm a rising senior at American High School in Fremont, California, headed toward a degree in Chemical Biology and Computer Science, while considering a minor in Applied Mathematics.
A little about me: I've been writing code since the seventh grade! It started with making games, which is still where a lot of my creative energy goes. As I went through high school, my interest in the physical sciences grew and shaped a lot of the projects I work on today. These days I spend most of my time in the space where chemistry, biology, and computing overlap: cheminformatics, small independent research projects, and the occasional hackathon.
Updated Summer 2026
Journal of High School Science (JHSS) · 2026
Journal of High School Science (JHSS) · 2026
Journal of High School Science (JHSS) · 2026
Journal of High School Science (JHSS) · 2026
Journal of High School Science (JHSS) · 2026
Regeneron Science Talent Search · in preparation · 2026
Independent research, mostly computational. Several more papers are in preparation; the code for every project is on GitHub.
A study of grokking and generalization that asks how the geometry of structured label noise decides how efficiently a limited audit budget can relabel data and recover clean performance.
A network-science look at the Online Encyclopedia of Integer Sequences, modeling sequences and their cross-references as a graph to surface hidden structure across mathematics.
Training sparse autoencoders on a chemistry language model (ChemBERTa) to pull apart the features it learns, a look at interpretability where chemistry meets machine learning.
Data, training, and mechanistic analysis of a small transformer that learns to classify aromaticity. It belongs to a series of interpretability probes into chemistry rules, alongside Woodward–Hoffmann and codon structure.
A run at the classic Traveling Salesman Problem, mentored by Aniket Sadashiva during my UC Berkeley internship, where I picked up a lot of dynamic programming.
A mobile app that helps seniors keep track of and remember their medications, built from scratch for the 2025 Congressional App Challenge.
A chemistry capstone at the intersection of computation and chemistry, with the code and a writeup coming here and to GitHub soon.
A MediaPipe app that reads external symptoms to flag possible disease early, before it becomes harder to treat.
An app to help overloaded students stay on top of school and their workload.
A couple dozen more — games, ML experiments, and early web builds — live on GitHub. See all on GitHub ↗
My portfolio! 12+ games built with and for the American High community through the Game Development Club, plus game jam entries.
Sitting / awaiting scores: Physics C: Mechanics, Physics C: Electricity & Magnetism, Macroeconomics, U.S. Government, Computer Science Principles, Biology, Statistics.