Building agent-based user simulator for multi-turn agent evaluation.
Applied Scientist Intern
Amazon
Designed and implemented a memory-augmented, multi-modal LLM agent-based framework for phishing detection.
Applied Scientist Intern
Amazon
Developed a diffusion model-based account takeover library to mitigate catastrophic forgetting, integrating contrastive loss and a tabular transformer encoder.
Education
PhD Computer Science
Purdue University
My research interests lie in LLM-red teaming, RL for security, LLM agents, and backdoor attack/defense.
MSc Electrical and Computer Engineering
Carnegie Mellon University
My research focused on explainable reinforcement learning.