About Me
Hi! I'm Wenkai Wang, a master's student at AI4GC Lab, Zhejiang University, advised by Shengyu Zhang.
My research focuses on building and evaluating computer-use agents that can operate real software through CLI and GUI interfaces. I am especially interested in agentic reinforcement learning for grounding and planning, as well as benchmarks that capture long-horizon professional workflows and human-in-the-loop collaboration.
Open to collaborations on CUA agents, agentic RL, and LLM evaluation.
Research Interests
- CLI/GUI Agent — Grounding, planning, and benchmark design for agents that interact with desktop and application UIs in realistic workflows.
- Agentic Reinforcement Learning — RL pipelines for post-training LLMs/MLLMs, including reward design, data synthesis, and stable self-improvement.
- LLM Evaluation — Building rigorous benchmarks and evaluation protocols for long-horizon agent behavior and multimodal reasoning.
Education
- 2025 – present · M.S. in Artificial Intelligence, Zhejiang University
- 2021 – 2025 · B.S. in Computer Science and Technology, Huazhong University of Science and Technology
Selected Papers
arXiv2026
DeskCraft: Benchmarking Desktop Agents on Professional Workflows and Human-in-the-Loop Collaboration
AAAI2026
Oral
A Rolling Stone Gathers No Moss: Adaptive Policy Optimization for Stable Self-Evaluation in Large Multimodal Models
arXiv2026