Research Intern
Alibaba · Qwen Agent Post-Training GroupWorking on post-training for agent capabilities in Qwen mainline models, with a focus on computer-use task testing and software-use workflows.
A research profile grounded in real software interaction.
I am interested in the moment when agent reasoning has to meet practical interfaces: messy software, professional tools, shifting workflows, and the need for reliable evaluation.
Research direction
GUI agents, LLM agents, agentic RL, and broader general-purpose agent systems.
My research sits at the intersection of multimodal reasoning, real-environment evaluation, and agent interaction systems. I am especially interested in how agent capabilities can scale from software interfaces toward more general task competence in the real world.
Across recent projects and internships, I have worked on plug-and-play improvement methods for GUI agents, real-environment benchmarks, large-scale trajectory data construction, multi-app desktop automation, and post-training pipelines for agent behavior.
My current work focuses on making agent systems more dependable in real environments by combining method design, benchmark construction, and systems-oriented implementation.
Interests
Topics I am currently excited about.
GUI agentsLLM agentsAgentic RLAGI
Timeline
Education and research experience.
PhD Student
Shanghai Jiao Tong University · X-LANCE LabResearching multimodal agents with an emphasis on GUI interaction, domain adaptation, and real-environment evaluation.
Research Intern
Beijing Institute for General Artificial Intelligence (BIGAI)Worked on multi-app macOS agents and early GUI-agent reinforcement learning pipelines, including desktop automation, task generation, and DAPO-based training adaptation.
Research Collaborator
Shanghai Jiao Tong University / BIGAIContributed to large-scale trajectory data construction and mobile GUI evaluation benchmarks.
B.Eng. in Computer Science
Shanghai Jiao Tong UniversityBuilt a strong foundation in systems, algorithms, architecture, and machine learning while moving into agent research.
Contact