I am a second-year Ph.D. student in Computer Science at the University of North Carolina at Charlotte, supervised by Prof. Haopeng Zhang. I spent the first year of my Ph.D. in Hawaiʻi 🌴. My research focuses on language-model agents: how to evaluate their behavior, design practical harnesses, and make them safer and more reliable in real-world workflows.

I am especially interested in contextual privacy, agent evaluation, and policy-grounded safety alignment.

I previously interned at Zoom in Bellevue, WA, working on agent safety and coding-agent harnesses, and at Microsoft Research Asia in Beijing, China, studying contextual privacy and evaluating agents in realistic workflows. See the Experience section for more details.

News

  • Celebration MPCI-Bench has been accepted to the NeurIPS 2026 Evaluations & Datasets Track.
  • Our paper An Empirical Study of Harness Design for Coding Agents is available on arXiv.
  • Started my Ph.D. in Computer Science at the University of North Carolina at Charlotte.
  • Completed a research internship at Zoom Video Communications, working on agentic safety and coding-agent harnesses.
  • Celebration Privacy in Action appeared in Findings of EMNLP 2025.

Research interests

Language-model agents and harnesses

I study how planning, tool interfaces, and context management affect the accuracy, cost, and reliability of long-horizon agents.

Contextual privacy and safety

I develop evaluations and mitigations for agents that work with sensitive information, with an emphasis on contextual integrity in multimodal settings.

Evaluation and alignment

I am interested in turning high-level safety principles into concrete behavioral rules, benchmarks, and diagnostic tools for deployed agents.

Selected publications

All publications →
  1. Coding harness study overview with context management, planning, and action-space ablations.
  2. AgentPrivArena connects seeded service records to an agent that reads information and executes actions.
  3. MPCI-Bench construction pipeline: image-based seed pairs, story refinement, and agent trace simulation.
  4. PrivacyChecker integration in an agent workflow using MCP tools and agent-to-agent communication.

Experience

CV →

  1. University of North Carolina at Charlotte

    Aug 2026 — Present

    Ph.D. Student in Computer Science

  2. Zoom Video Communications, Bellevue, WA

    May 2026 — Aug 2026

    Research Intern — Agentic Safety / Agent Harness

  3. University of Hawai‘i at Mānoa

    Sep 2025 — Aug 2026

    Ph.D. Studies in Computer Science; Research Assistant — Contextual-Integrity Privacy & Alignment

    • Developed MPCI-Bench, a benchmark for evaluating contextual-integrity behavior in multimodal language-model agents.

  4. Microsoft Research Asia, Beijing, China

    Dec 2024 — Sep 2025

    Research Intern — Social Computing & DKI Groups

    • Built PrivacyLens-Live, a realistic MCP/A2A-based benchmark environment for LLM agents; first author of Privacy in Action.
    • Shipped a production meeting-time recommender for Microsoft Teams.

  5. Wuhan University, Hongyi Honor College

    Sep 2021 — Jun 2025

    B.Eng. in Computer Science

  6. Beijing Capital Eco-Environment Protection Group, Beijing, China

    Oct 2023 — Aug 2024

    Developer — Knowledge-Based LLM Chatbot

    • Built a production RAG pipeline over the company knowledge base for an internal Q&A chatbot.

Blog

All posts →

I plan to share occasional notes on language-model agents, research practice, and graduate school.