About Me
I am a Ph.D. student in Computer Science at the University of Virginia, advised by Prof. Aidong Zhang. Prior to this, I received my Bachelor’s degree in English Language from Tsinghua University. My research focuses on building LLM and multimodal agents that search and reason over long horizons. I optimize how these agents acquire and use information through both post-training (e.g., reinforcement learning) and training-free methods (e.g., in-context learning). I also investigate how these models work internally and how to evaluate them, with the goal of making them faithful, interpretable, and controllable. More details can be found in my CV.
I am actively seeking full-time industry opportunities starting in 2027. Please feel free to reach out!
News
- Oct 2026: One paper is accepted to Nature Communications!
- Sep 2026: Three papers are accepted to NeurIPS 2026!
- Aug 2026: One paper is accepted to EMNLP 2026!
- Jun 2026: One paper is accepted to ECCV 2026!
- May 2026: Start my research scientist internship at Meta!
- May 2026: One paper is accepted to KDD 2026!
- Feb 2026: One paper is accepted to CVPR 2026!
- Jan 2026: One paper is accepted to ICLR 2026!
- Jan 2026: One paper is accepted to AISTATS 2026!
- Jan 2026: Our MedRAG toolkit reaches 500 stars on GitHub!
- Nov 2025: One paper is accepted to AAAI 2026!
- Aug 2025: One paper is accepted to EMNLP 2025!
- Jun 2025: One paper is accepted to ICCV 2025!
- May 2025: Start my applied scientist internship at Amazon!
- May 2025: Two papers are accepted to IJCAI 2025!
Research Interests
- Agents, RAG & Post-training: agentic search, retrieval-augmented generation, long-term multimodal memory, and long-horizon reasoning, optimized through both post-training (e.g., reinforcement learning) and training-free methods (e.g., in-context learning). Representative papers: MemLife, RAG-Gym, CIRCLES, MedRAG.
- Faithfulness & Evaluation: hallucination detection, attribution faithfulness, and benchmarks for LLMs and VLMs. Representative papers: RAGLens, MedFocus, TruthHypo.
- Interpretability & Control: model internals, mechanisms of in-context learning, and interpretable models. Representative papers: Task Operators, Neural Additive Experts, ProtoNAM.
My work has broad applications, from healthcare and scientific discovery to wearable assistants and personalized shopping. For a complete list of papers, see the Publications page.