Education

B.S. in Computer Science

University of California, Santa Barbara
06/2027

Relevant Courses: Foundation Models (grad), Machine Learning, Generative AI, Operating Systems, Parallel Computing

GPA: 3.98

Experience

ChipAgents | AI for Chip Design

ML Engineer Intern
03/2026 – Present | Santa Barbara, CA
  • •Developed and shipped core agent capabilities for long-horizon hardware workflows, including async subagents, agent teams, background bashes, and user steering.
  • •Building a benchmark to evaluate ChipAgents performance on functional coverage, diagnosing failure modes and improving performance on real-world SoC repositories.

    • UCSB NLP Group⁠

    Undergraduate Researcher | Advisor: Xin Eric Wang
    10/2025 – Present | Santa Barbara, CA
  • •Proposed an evolving benchmark and task generation method to challenge frontier models on multi-agent cooperation and theory of mind, exposing failures in safe and trustworthy AI.
  • •Released an open-source evaluation framework⁠ and task library on top of the Meta Habitat robot simulator supporting 10+ agents in a realistic household, enabling future directions in multi-agent cooperation research.
  • •Developing a testbed to understand long-context degradation in coding agents and foundation models, isolating context rot from task difficulty.
    • 06/2025 – 09/2025 | San Jose, CA
  • •Wrote depthwise convolution kernels in Rust to achieve 16x faster performance on our NPU. Evaluating on MobileNetV2 demonstrated a 5.8x improvement in inference latency.
  • •Explored efficient autoscheduling for CNN/​MLP operations by extending Halide to support in-memory compute primitives.
  • •Performed INT8 post-training quantization (PTQ) on CNN-based vision models with PyTorch, targeting our NPU release.
    • Research

    Context Rot Bench

    In Progress
    05/2026

    Why do coding agents degrade over long-horizon context and how can we improve existing harnesses?

    EnactToM: An Evolving Benchmark for Embodied Theory of Mind

    Under Review, NeurIPS
    05/2026

    Benchmarking frontier agents on functional theory of mind (ToM), the ability to interpret and act on other's beliefs.

    Awards
    10/2025

    Won the world's largest collegiate hackathon (3k+ participants, 700+ teams) building a computer-use agent platform to remotely control Macs through iMessage and FaceTime. Released 1 month before OpenClaw.

    Projects
  • •Trained a Mixture-of-Experts DQN agent achieving >90% human-level completion rate on Geometry Dash levels
  • •Developed an OpenAI Gym pipeline to evaluate different models and RL algorithms, culminating in a technical report
  • •Built a C++ Geometry Dash mod for real-time exchange of screenshots and actions between the AI agent and game