👤 About

I am Yichen Pan(潘奕琛), a first-year Master’s student of the Data Science Program at Zhejiang University. My main areas of research are Large Language Models (LLM) and Natural Language Processing (NLP). Currently, I am focused on projects involving the LLM Agent and the RAG to enhance the capabilities of LLMs.

Feel free to contact me via my email at panyichen@zju.edu.cn or visit my GitHub page at github.com/pan-2001 to learn more about my research and projects, or for academic and technical discussions.

---

🔥 News

[2024-08]  WebCanvas has been accepted by the ACL NLRSE Workshop. I am honored to have been selected for an oral presentation. I look forward to meeting and exchanging ideas with you in Bangkok!

[2024-06]  WebCanvas has been accepted by the ICML TiFA Workshop and I am fortunate to have been selected for a Paper Lightning Talk!
Click to see the poster.

[2024-06]  My latest work, WebCanvas, an innovative framework for benchmarking web agents in online environments, is now available on arXiv and GitHub. Join us and explore the future of web agent evaluation!

---

📖 Publications

* indicates equal contribution

WebCanvas: Benchmarking Web Agents in Online Environments
Yichen Pan*, Dehan Kong*, Sida Zhou*, Cheng Cui*, Yifei Leng, Bing Jiang, Hangyu Liu, Yanyi Shang, Shuyan Zhou, Tongshuang Wu, Zhengyang Wu
Preprint, 2024
[Paper][Platform][GitHub][Poster]

---

🎓 Education

Master’s Student in Data Science Program, Zhejiang University(ZJU)
2023 - Present
Advisor: Xianliang Hu

Undergraduate, University of Science and Technology Beijing(USTB)
2019 - 2023

---

🖥️ Internship Experience

SUPCON Technology Co., Ltd.  中控技术股份有限公司
2024.6 - Present
- Currently doing LLM RAG and document QA job.

iMean AI, Stellarrover Inc.  跨越星空科技有限公司
2023.7 - 2024.5
- Authored the “WebCanvas” paper, currently available on Arxiv.
- Established WebCanvas framework for Web Agents’ online evaluation.
- Created Mind2Web-Live dataset for use with WebCanvas.
- Developed WebCanvas Agent for autonomous web navigation.
- Integrated zero-shot web navigation capability into iMean Copilot.
Advisor: Dehan Kong

SuperSymmetry Technologies  超对称科技有限公司
2022.12 - 2023.5
- Participated in the development of BBT 2.0 LLM.
- Pretrained models using DeepSpeed and Megatron on the Sugon cluster.
- Fine-tuned BBT 2.0 using full parameters and LoRA on 8*A100 GPUs.
- Accelerated LLM inference speed by using parallelism and quantization.
- Conducted benchmark tests for the model using C-Eval, CLUE, etc.

Institute of Software, CAS  中国科学院软件研究所
2022.3 - 2022.7
- Developed the “Go-sqlancer” testing system, implemented in Go.
- Generated SQL queries and splits them into three sub-queries using TLP.
- Evaluated DBMSs by comparing results from the original and subqueries.
Advisor: Wensheng Dou