Systems and studies that shaped my current research agenda.

A plug-and-play framework for reducing planning and grounding bias in domain-specific GUI agents by retrieving and distilling task-relevant tutorial knowledge.

A real-environment benchmark for multimodal agents working with professional materials science software, covering GUI operation, code execution, and cross-tool workflows.

A large-scale trajectory data construction effort that transforms multimodal web tutorials into agent training data across operating systems and application types.

A multi-application desktop agent for macOS workflows, including browser use, calendar management, and messaging coordination.

Early-stage reinforcement learning work for GUI agents, including task generation and training-method adaptation.

A benchmark effort for evaluating LLM-based GUI interaction in mobile environments with isolated tasks, replay infrastructure, and behavior analysis.