My research interests lie broadly in the realm of formal methods and their intersection with machine learning. Recently, my research has focused on utilizing formal methods to enhance the reliability and efficiency of LLMs, during both inference and training.

For examples, check out Diffinity, where we formally steer diffusion models to generate text that satisfies a regular expression, or Constraint-Aware Training, where we show how LLM training can be made more efficient by applying formal information that can be by inferred by an external symbolic engine.

I am on the job market this academic year (2026-2027), for positions in both academia and industry!