I build large language models end to end (from data curation and pretraining to fine-tuning and evaluation) with a focus on code generation in settings the field tends to overlook.
I am completing my Ph.D. in Computer Science at George Mason University, advised by Dr. Marcos Zampieri and Dr. Antonios Anastasopoulos. My dissertation, Exploring and Adapting Code LLMs for Underrepresented Domains, asks a recurring question: what happens to code models once you step outside English, outside Python, and outside the benchmarks everyone already optimizes for?
In Fall 2026 I join the University of Notre Dame as a Provost's Postdoctoral Fellow in Computer Science & Engineering, working with Dr. Joanna C. S. Santos on safety-by-construction guardrails and multilingual program synthesis for Code LLMs. Along the way I have released open benchmarks, corpora, and models, including mHumanEval, TigerLLM, and MojoBench, that the community can build on.
Open to collaboration
I am always glad to talk with people working on Code LLMs, multilingual NLP, or LLM safety. If you would like to collaborate, the fastest way to reach me is email.
Code LLMs & program synthesis
Adapting code models to low-resource programming languages and underexplored domains.
Multilingual & low-resource NLP
Evaluation and modeling across natural languages, with a focus on Bangla and code-mixed text.
LLM safety & AI in CS education
Guardrails for code assistants and the role of LLMs in introductory computing.
Benchmarks & datasets
Large, openly released resources that make under-tested settings measurable.