Jiarui Yan

MS in Computational Data Science, School of Computer Science, Carnegie Mellon University

prof_pic.jpg

Pittsburgh, PA

jerryy2 [at] cs.cmu.edu

I am a master’s student in Computational Data Science at Carnegie Mellon University. At CMU’s Language Technologies Institute I lead two projects: pre-training data curation with Chenyan Xiong (ReScraper) and long-horizon auto-research agents with Yiming Yang (TraceML, NeurIPS 2026). I have also worked on agentic systems with William W. Cohen (sponsored by Google DeepMind) and on predicting the RL utility of training environments with Sanmi Koyejo at Stanford.

My research asks what actually makes a language model better: which data is worth training on, which environments make RL pay off, and why agents still fail on long-horizon work.

This summer I was a software engineer intern at Google, building AI agents. I graduated magna cum laude from the University of Washington. I am looking for research scientist / engineer roles starting 2027.

news

Sep 28, 2026 Released ReScraper, one 0.6B model that replaces the heuristic scraper and filter stack for LLM pretraining data: the paper, a project page, code, the model, and the curated corpus.
Sep 24, 2026 TraceML was accepted to the NeurIPS 2026 Evaluations & Datasets Track and will also appear at the NeurIPS 2026 Workshop on Evaluation of Interactive Agents. See the paper, the project page, and the open-source toolkit for reading any ML-agent run against human Kaggle practice.
Jul 10, 2026 TraceML was accepted to the WAB workshop at COLM 2026.
May 11, 2026 Started my software engineering internship at Google in Sunnyvale, building AI agents.
Jan 12, 2026 Joined the ReScraper project: one small model that scrapes and cleans web data for LLM pretraining.

selected publications

  1. NeurIPS
    traceml.png
    TraceML: What Auto-Research Agents Miss in Long-Horizon ML Development
    Jiarui Yan*, Weiwei Sun*, Sijie Li, Wenhan Li, and Yiming Yang
    In Advances in Neural Information Processing Systems (NeurIPS), Track on Evaluations and Datasets, 2026
    Also at the COLM 2026 Workshop on Agent Behavior and the NeurIPS 2026 Workshop on Evaluation of Interactive Agents
    arXiv Code Website
  2. arXiv
    rescraper.png
    ReScraper: Unified Scraping and Cleaning of Web Data for Effective LLM Pretraining
    Zichun Yu*, Jiarui Yan*, Shlok Sanghvi, Nihar Atri, and Chenyan Xiong
    Under review, 2026
    arXiv Code Website
  3. arXiv
    agentic.png
    Learning to Construct Practical Agentic Systems
    Aditya Kumar*, Zhihan Lei*, Jiarui Yan*, Joshua W. Momo, Lauhitya Reddy, Rafael Enrique Cabrera Jimenez, Cassandra A. Cohen, Arthur Kajiyama, and William W. Cohen
    Under review, 2026
    arXiv