About Me

Greetings! I am an undergraduate at UW–Madison working on Health AI and data-centric AI. I am fortunate to be advised by Prof. Yin Li and Prof. Fred Sala. I have also gained valuable AI and ML engineering experience through internships at Freecharge, Frizzle, and UW–Madison Department of Physics.

My research focus is on efficient AI with an interest in health care AI as well. My past work was on weight-space analysis to recover the training data distribution. Currently, I am working on zeroth order optimization and on improving NICU mortality prediction.

News

  • May 2026Joined Frizzle (YC S25) as a Software Engineering Intern.
  • May 2026WARP accepted to the ICML 2026 Weight Symmetries Workshop.
  • Nov 2025Joined Yin Li's lab to work on NICU mortality forecasting.
  • Jul 2025Joined Sprocket Lab as an ML Research Engineering Intern.
  • May 2025Joined Freecharge as a Machine Learning Intern.
  • Nov 2024Joined UW–Madison Physics as a Software Engineering Intern.
  • Sep 2024Started as a UG Teaching Assistant for CS 540 and CS 300.
  • Sep 2023Started undergrad at UW–Madison in Computer Science and Data Science.

Projects

Publications

  • Illustration for WARP: Weight-Space Analysis for Recovering Training Data Portfolios

    WARP: Weight-Space Analysis for Recovering Training Data Portfolios

    Tzu-Heng Huang*, Aditya Goyal*, John Cooper, Frederic Sala

    ICML 2026 · Weight Symmetries Workshop

    Paper ↗Code ↗Poster ↗Media ↗

Research Experience

Yin Li Lab

Dec 2025–present

UG ML Researcher

Improving NICU mortality prediction over traditional statistical approaches by Developing a Transformer-based framework that leverages pre-trained time-series encoder representations.

Sprocket Lab

July 2025–present

UG ML Researcher

Co-developed WARP, a method that uses gradient signals to recover training data distributions from model weights, leading to a first-author paper at the ICML 2026 weight symmetries workshop. Currently leading research on reducing variance in zeroth-order optimization by better aligning the perturbation subspace with the true gradient.

SWE Experience

Frizzle(YC S25)

June 2026 - September 2026

Software Engineering Intern

Worked on agentic workflows and evals for AI-powered math grading at a Y Combinator-backed startup, owning features from workflow design through backend integration and implementation. Shipped three end-to-end grading pipelines: IGCSE rubric generation, scratch-paper grading, and IB paper grading.

Build strong evals, develop effective AI workflows that maintain momentum without sacrificing understanding, and deeply understand the problem to make better assumptions and design decisions.

Freecharge

May 2025 - August 2025

Machine Learning Intern

Built AI agents for financial analytics using AWS Bedrock and LangGraph. Developed a financial search engine over 12,000+ web pages with AWS Kendra, along with YOLOv8-based video analytics for CCTV footage.

Think in production instead of local, quantify decisions with metrics and automate wherever possible.

UW–Madison Physics

Nov 2024– Feb 25

Software Engineering Intern

Built a framework to train MLPs on mathematical functions and visualize activation flow across layers using PCA

Follow strong engineering practices and and communicate the purpose of your work in a way that matches your audience’s technical background.

Writing

Teaching

  • CS 300 — Programming II

    UG Teaching Assistant · 2024–present

  • CS 540 — Introduction to Artificial Intelligence

    UG Teaching Assistant · 2024–25

Outside Research

When I’m away from a computer, you’ll usually find me playing squash or underwater.

Green sea turtle on a sandy reefScuba diving, seated mid-water
Banded sea krait above the sand

Scuba Diving

PADI Open Water diver, working my way towards Advanced and Rescue. I love scuba because it’s one of the few places you can see wildlife truly in the wild, even if that means being two metres from a sea snake. Only complaint: the water’s salty.

Playing squash

Squash

Not the vegetable. The actual sport. There isn’t a game I enjoy more or another I can play for three hours straight. Fast, intense, and constantly making you think on the spot.