Project Lead
AgentDebugX
The first comprehensive debugging framework for long-horizon agent trajectories, turning failed runs into auditable diagnoses, repairs, and verifiable reruns.
Selected Research & Engineering
Projects
Open-source agent systems, multimodal research, and responsible-AI tooling built for real-world use and reproducible research.
GitHub Profile
Reason · Act · Improve
Agent Systems
Frameworks for capable, self-improving, and tool-using AI agents. Core Author
OpenManus & OpenManus-RL
Core author of OpenManus-RL in an open-source agent ecosystem with 62,000+ combined GitHub stars.
See · Structure · Reason
Multimodal Intelligence
Research systems connecting visual information with language-model reasoning. Research Project
SeeingEye
Agentic information-flow framework that unlocks multimodal reasoning for text-only LLMs.
Measure · Diagnose · Mitigate
Responsible AI
Methods for identifying failure modes and reducing harmful model behavior. Researcher
Mitigating Bias Inheritance in LLM Data Augmentation
An automated experimentation pipeline for measuring and mitigating how bias propagates through LLM-generated synthetic data.