About

Welcome!

I am a Machine Learning Scientist at the Neuron Science team in Amazon Annapurna Labs, working on the design and development of large-scale training algorithms. I received my Ph.D. degree in Electrical and Computer Engineering from UCLA, advised by Prof. Suhas Diggavi. Prior to UCLA, I obtained my Bachelor’s degree in Electrical and Electronics Engineering (with a Minor in Economics) from Bilkent University.

I design algorithms that make large-scale machine learning faster, more reliable, and more useful; whether the goal is training a model at scale or adapting one to the constraints and preferences of an individual user. Concretely, my work is organized around three threads:

LLM training efficiency — system-aware optimizers and low-precision training recipes (e.g., FP8, stochastic rounding) that scale to billion-parameter regimes without sacrificing convergence.

Post-training for kernel optimization and code generation — reinforcement learning and fine-tuning techniques that turn LLMs into effective tools for generating high-performance code and kernels.

Personalized and efficient distributed machine learning — personalized federated learning, model compression, and privacy-preserving algorithms for settings where data, compute, or trust are heterogeneous across clients.

A common thread across these directions is the interplay between optimization and practical algorithm design: I am drawn to problems where the right answer comes from understanding the loss landscape, the data distribution, and the system together; and where the resulting algorithm makes machine learning meaningfully more efficient or more personalized for the people who use it.