Research
Stable Signer
A message-conditioned watermarking framework for image provenance. The design balances model providers’ control over their models, a third party’s need for a shared extraction interface, and users’ exposure to quality loss and false attribution.
I contributed to the framework design, implementation, experiments, and analysis at Zhejiang University. The work is not yet published.
Agent safety
As a research assistant at NUS (2024–2025), I worked on safety in multi-step LLM and agent workflows with Prof. Zhenkai Liang at the CuriOSity Research Group.