Research Experience

Research Assistant
Jun 2024 – Aug 2025
Prof. Yushun Dong's Group, Florida State University
  • Contributed as a principal contributor and maintainer to the PyGIP library, which consolidates code and datasets from research papers on GNN model extraction attacks and defenses.
  • Developed the defense framework and implemented 2 model extraction defense algorithms and 8 model extraction attack algorithms into library, enabling users to select attack models from one paper to test against defense models from another. This allows for flexible and comprehensive benchmarking of GNN security methods.
  • Co-authored the PyGIP 1.0.0 documentation, a comprehensive guide for users on installation, API reference, and usage of the library.
Research Assistant
May 2026 – Now
Research Assistant in Prof. Xinyu Wang’s Group, University of Michigan-Ann Arbor
  • Developed VeriEQL, a formal verification framework in Lean 4 that models SQL semantics and proves query equivalence through constraint-aware relational algebra reasoning.
  • Constructed a library of verified SQL transformation lemmas to support automated equivalence proofs for query rewrites involving joins, aggregation, subqueries, and NULL semantics
  • Investigated LLM-based proof generation by fine-tuning code models with Lean-verified proof traces, demonstrating the interaction between symbolic reasoning systems and neural theorem provers for scalable SQL verification
  • Developed an induction-based proof workflow combining symbolic goal decomposition and lemma application with agent-driven lemma generation, focusing AI reasoning on local proof obligations while ensuring that resulting proofs are checked by Lean 4
Summer School
Jun 2026 – July 2026
Oregon Programming Languages Summer School (OPLSS) 2026, University of Oregon, Eugene
  • Selected participant in an intensive graduate-level program focused on the intersection of program verification, type theory, and semantics.
  • Completed advanced coursework in Separation Logic (D. Dreyer), Rocq/Coq Proof Assistant (A. Azevedo de Amorim), and Modal Logic (E. Pimentel)
  • Developed practical skills in interactive theorem proving (Rocq) and auto-active verification (R. Leino), with exposure to state-of-the-art research on climate-resilient computing infrastructure.
Research Assistant
Jun 2025 – Dec 2025
Prof. Remzi Arpaci-Dusseau's Group, University of Wisconsin-Madison
  • Reproduced more than 50 semantic bugs and tested the severity of each semantical bug about scheduler from Linux, Kubernetes and Volcano.
  • Classified these bugs based on 6 different criteria: special input, source of violation, root cause, influenced decision, how the bug is found, and after fix.
  • Co-developing a bug testing tool.