Research

Research Assistant, University of Utah · May – Sep 2026
Advisors: Dr. Yue Jiang and Dr. Ben Greenman

Detecting accessibility failures that emerge during interaction, beyond what a static page check can reveal.

Developed a taxonomy of navigation failures affecting blind and low-vision screen-reader users, drawing on prior literature and user studies.
Formalized navigation failures as Linear Temporal Logic (LTL) properties and implemented checks over interface interactions.
Built a web-agent baseline that navigates through screen-reader feedback and keyboard-only interaction.
Evaluated the checker on generated websites, achieving 88.7% precision.

Improving Generative UIs from Use

Research Assistant, Purdue University · May 2026 – Present
Advisor: Dr. Jason Wu

Studying how interaction traces can help AI-generated interfaces become more usable.

Built a pipeline to generate websites, collect task interaction traces, diagnose usability issues, and repair interfaces with coding agents.
Used trained computer-use agents as baselines for simulating user behavior and providing usability feedback.
Investigating PPO-based fine-tuning and preference optimization using interaction feedback, with controlled experiments on task usability.

WebCoEvo: Adversarial Co-Evolution for Web Agents

Research Assistant, University of Michigan · Mar – Jun 2026
Advisor: Prof. Honglak Lee

Helping web agents transfer what they learn as websites and interfaces change.

Built an adversarial co-evolution framework that pairs a coding-agent-driven UI drift generator with a web agent.
Extracted generalizable reflection rules from agent failures and compared them with an ExpeL baseline.
Worked on a knowledge-graph pipeline to identify tasks affected by website version changes.
Containerized multiple website versions with Docker and evaluated out-of-distribution generalization using BrowserGym and AgentLab.

Gravitational Effects on Microorganism Swarming

Research Assistant, Shanghai Jiao Tong University · Sep 2024 – Aug 2025
Advisor: Dr. Zijie Qu

Automating colony boundary segmentation for the study of microorganism swarming.

Trained a U-Net for colony boundary segmentation and diagnosed systematic failure modes.
Adopted a zero-shot Segment Anything Model (SAM) pipeline for reliable automated detection, replacing a manual annotation pipeline.