About
Hi, I’m Ethan Li, a Computer Science undergraduate at The Hong Kong Polytechnic University. I previously worked as an undergraduate research assistant with the PolyU NLP Group and am currently interning with Tencent PCG’s Machine Learning Platform team.
What I work on
My current work centers on AI agents: how to evaluate them beyond a single benchmark score, how agent harnesses and infrastructure shape their reliability and capability, and how coding agents can be built into genuinely useful systems.
I am particularly interested in three connected directions:
- Agent evaluation: Evaluating judgment, trajectories, and failure patterns beyond benchmark scores and across real-world tasks.
- Agent harnesses and infrastructure: Building context, memory, skills, tools, sandboxes, and runtimes that make agents reliable and effective.
- Coding agents: Designing software-agent loops that can plan, execute, verify, and evolve with human or expert oversight.
Research and engineering
I tend to work where research meets engineering. I enjoy turning ideas into systems that can be run, observed, and tested, rather than leaving them as isolated prototypes.
At Tencent PCG, I work on agent systems and model infrastructure, focusing on reliability, evaluation, performance optimization, and bringing AI capabilities into production.
I also build AgentCorp, a loop-engineering system for software delivery. Through skills and quality gates spanning requirements, design, implementation, review, testing, and acceptance, it makes agent work controllable, understandable, and verifiable across both Claude Code and Codex.
Previously, at Tencent Security Xuanwu Lab, I worked on SecEval, a benchmark for evaluating foundation models’ cybersecurity knowledge with more than 2,000 questions across nine security domains.