Hi, I’m Xinyu Jessica Wang (王昕雨)!
I’m a second-year PhD student in Computer Science at the University of Wisconsin–Madison. My research asks how to build AI agents that can reason and act reliably in complex, real-world environments.
I am especially interested in capabilities that transfer across scientific, embodied, and social settings, and in agents that remain useful, understandable, and aligned with human needs over long horizons.
I am currently a research intern with Snowflake AI Research, where I work on long-horizon agent optimization, including evaluation, agent harnesses, and post-training.
I am fortunate to work with Prof. Bilge Mutlu and Prof. Robert Nowak, with mentorship from Christine Lee and Haoyue Bai.
CV · Résumé · Google Scholar
Research Interests
- Long-horizon agents: optimization, evaluation, harness design, and post-training for extended tasks.
- Grounded intelligence: agent reasoning across scientific, embodied, and social environments.
- Human-centered AI: systems that augment people while remaining understandable and responsive to human intent.
Publications
- LearnMate²: Design and Evaluation of an LLM-powered Personalized and Adaptive Support System for Online Learning. DIS ‘26. Paper · arXiv
- U-Define: Designing User Workflows for Hard and Soft Constraints in LLM-Based Planning. arXiv preprint. Paper
- The Long-Horizon Task Mirage? Diagnosing Where and Why Agentic Systems Break. IAEval @ NeurIPS 2026 — Oral. Paper · Project · Workshop
- LearnMate: Enhancing Online Education with LLM-Powered Personalized Learning Plans and Support. CHI EA ‘25. Paper
- VeriPlan: Integrating Formal Verification and LLMs into End-User Planning. CHI ‘25. Paper
News
- Sep 2026: Agents’ Last Exam was accepted to the NeurIPS 2026 Evaluations & Datasets Track.
- 2026: The Long-Horizon Task Mirage? was selected for an oral presentation at the IAEval Workshop at NeurIPS 2026.
- Jun 2026: LearnMate² was published at DIS ‘26.
- Apr 2025: Presented LearnMate at CHI ‘25 in Yokohama, Japan. Poster details · Paper