Sairam Vaidya

2nd Year Masters Student at UCSD

About

I am a second-year graduate student at UCSD working with Loris D'Antoni. My research focuses on constrained generation for language models and their applications to compiler testing and program synthesis. Previously, I worked on low-latency trading systems at J.P. Morgan and Morgan Stanley. I received my BE in Computer Science from PSG Tech in 2023.

Publications

Constrained Sampling for Language Models Should Be Easy: An MCMC Perspective

Emmanuel Anaya Gonzalez*, Sairam Vaidya*, Kanghee Park, Ruyi Ji, Taylor Berg-Kirkpatrick, Loris D'Antoni

NeurIPS 2026

arXiv →

Constrained Adaptive Rejection Sampling

Paweł Parys, Sairam Vaidya, Taylor Berg-Kirkpatrick, Loris D'Antoni

Preprint

arXiv →

Projects

Kestrel

Semantic fuzzing framework for MLIR compiler dialects using LLM-based test generation with grammar-constrained decoding.

GitHub →

CASA

Python package implementing constrained sampling algorithms (CARS, MCMC, rejection sampling variants) for language models.

GitHub →

SemGuS-LENS

LLM-guided program synthesis framework combining probabilistic enumeration with iterative PCFG refinement for semantic-guided synthesis.

GitHub →

Convex IRL

Convex formulation of inverse reinforcement learning with global optimality guarantees and robustness to noisy expert demonstrations.

GitHub →

Teaching

CSE 20: Discrete Mathematics

Teaching Assistant, Fall 2025

University of California, San Diego

CSE 20: Discrete Mathematics

Teaching Assistant, Summer 2025

University of California, San Diego

Fun Facts

  • Favorite Pokémon: Lapras, Umbreon and Serperior
  • Chelsea will win Champions League this year