About me

I'm an ML researcher and engineer working on making large-model training cheaper, faster, and more stable — through low-precision (FP8 / FP4) arithmetic, custom GPU kernels, and efficient distributed training on multi-node, multi-GPU clusters.

At Huawei Canada, as an Efficient AI Researcher, I work on low-bit training, model pruning, and test-time adaptation, write kernels in Triton and CUDA, and pre-train large transformers at scale. At Mila – Quebec AI Institute and Université de Montréal, I do research in Professor Irina Rish's CERC-AAI Lab on foundation models for time-series and neuroimaging data.

My path started with hardware. I wrote my first C++ program in middle school, and a love for mathematics pulled me toward algorithms and competitive programming (ACM-ICPC and IEEEXtreme). While studying Computer Engineering at Isfahan University of Technology, I realized that what I enjoy most is not designing hardware but understanding it deeply — and then using it efficiently with the right algorithms.

That led me to an M.Sc. in Algorithms and Computation Theory at Shahid Beheshti University, where my thesis brought algorithms to bioinformatics (AFITbin, published in BMC Bioinformatics), and then to an M.Sc. in Artificial Intelligence at Université de Montréal. Today the same instinct — hardware awareness plus good algorithms — drives my work on efficient deep learning.