Junye Pan

Email: junyep2@illinois.edu  |  GitHub: Junye-Pan  |  LinkedIn: Junye Pan

I am currently pursuing a Master of Computer Science degree at UIUC, after earning a B.S. in Computer Science and a B.A. in Mathematics from Boston College. My current interests center on AI for coding and software engineering agents, with a focus on building AI systems that can reason, generate code, and improve their performance through structured feedback.

My recent work focuses on SWE agents, AI for coding, and agentic optimization. I am particularly interested in how AI systems can use structured feedback and additional computation to improve their performance on software engineering, mathematical reasoning, and systems optimization tasks. My broader experience includes agentic systems, retrieval-augmented generation, and representation learning.

Portrait of Junye Pan

Interests

  • AI for coding and software engineering agents
  • Retrieval-augmented generation and tool-using AI systems

Works & Projects

  • A Framework for Agentic Optimization [report] [code]
    A server-first framework for letting coding agents conduct optimization search while benchmark authority stays under server control.
  • Coding-agent trajectories passing through a verifier
    Weak-to-Strong Temporal Verification for Coding Agents [report]
    A trajectory-aware verifier for selecting stronger SWE-agent patches from multiple candidate runs using weaker judge models.