I am an undergraduate student at Tongji University, advised by Prof. Miaomiao Zhang.
Since April 2025, I have been a visiting student at Westlake University, advised Prof. Yue Zhang. I also work closely with Prof. Qi Zhu at Zhejiang University, where I am researching the attention mechanisms of Vision-Language Models and working to enhance their long-context reasoning abilities. Since April 2026, I have been a research intern at the University of Virginia, advised by Prof. Yu Meng. My research focuses on post-training of LLM.
🔥 News
- 2026.07: 🎉 Our paper “DeepInstructor: An Agentic AI Instructor for Experience-Driven Idea Evaluation” has been accepted by COLM 2026 Workshop LM4Sci!
- 2026.05: 🎉 Our paper “VERA: Identifying and Leveraging Visual Evidence Retrieval Heads in Long-Context Understanding” has been accepted by IJCAI 2026!
- 2026.04: 🎉 Our paper “Augmenting Multi-Technique Static Analysis with Large Language Models” has been accepted by ISSTA 2026!
🌳 My Experience Tree
I am committed to comprehensively understanding and researching the full pipeline of LLM training and inference. Below is my experience tree.
📝 Publications
VERA: Identifying and Leveraging Visual Evidence Retrieval Heads in Long-Context Understanding
Rongcan Pei, Huan Li, Fang Guo, Qi Zhu
- Through attention analysis, we identify specific Visual Evidence Retrieval (VER) Heads — a sparse, dynamic set of attention heads critical for locating visual cues during reasoning. We propose VERA, a training-free framework that detects model uncertainty to trigger explicit verbalization of visual evidence. VERA achieves 21.3% relative improvement on Qwen3-VL-8B-Instruct and 20.1% on GLM-4.1V-Thinking across five benchmarks.
DeepInstructor: An Agentic AI Instructor for Experience-Driven Idea Evaluation
Rongcan Pei, Fang Guo, Qinglin Qi, Qi Zhu, Yun Luo, Jianhao Yan, Minjun Zhu, Qiujie Xie, Dehong Zheng, Yue Zhang
- We propose DeepInstructor, the first agentic AI instructor that emulates how human mentors leverage past scholarly experience to guide research ideation. DeepInstructor extracts experience triples from large-scale review corpora, constructs an Experience Graph, and employs an agentic reasoning framework for experience-driven idea evaluation.
Qi Zhu, Fang Guo, Rongcan Pei*, Shuqi He, Hui Chen, Yue Zhang
- An intelligent paper recommendation and digestion platform that adopts a temporally decoupled architecture. The system autonomously fetches, indexes, and digests newly released arXiv papers through a daily pipeline, with multimodal LLM-based summarization executed offline.
Junxiang Wang, Fu Song, Miaomiao Zhang, Bowen Du, Rongcan Pei
- We propose ELSA, a neuro-symbolic approach that augments static analysis with LLM-assisted constraint-guided reasoning and resolves conflicting outputs through analyzer ensemble. ELSA significantly outperforms baselines across three open-source benchmarks and self-constructed ZKP-based smart contracts.
🎖 Honors and Awards
- 2025, Social Activity Scholarship, Tongji University.
- 2024, Guo Xie Birong Scholarship (equivalent to University First-Class Scholarship), Tongji University.
📖 Educations
- 2023.09 - Present, Undergraduate Student, Tongji University. Advised by Prof. Miaomiao Zhang.
💻 Service
- Invited as reviewer for KDD 2026.