Research

Research Experience

  • Agentic Compiler Testing for ML Compilers (2026-Present)
    • At Amazon Annapurna Labs, develop agentic compiler testing for semantic miscompilations in ML compilers.
    • Found 120+ bugs in TileLang (50+ fixed) and about 10 bugs in Triton.
  • Scalable and Incremental Probabilistic Logic Inference (2025-Present)
    • Develop scalable and incremental probabilistic logic inference with Soufflé and ProbLog, with applications to program analysis.
    • Incremental Inference for Probabilistic Datalog has been accepted to CAV 2026.
    • PSoufflé: Scaling Exact Probabilistic Logic Inference for Program Analysis has been accepted to FMCAD 2026.
  • Verified Polyhedral Scheduling (2022-2024)
    • Built a Coq-verified validator for affine polyhedral schedules.
    • Validated Pluto schedules against CompCert semantics; released it as PolCert.
    • Published at TASE 2024.
  • Verified Optimization under Promising Semantics (2021-2022)
    • Verified common-subexpression elimination in Coq under promising semantics.
    • Released as part of a PLDI 2022 artifact.