Research Scientist
LG AI Research, Ann Arbor, MI · Ph.D., Stanford University
I build large language model systems that are more capable, reliable, and efficient, spanning LLM agents, high-quality data generation, and scalable optimization for large models.
Integrating large-scale language models with autonomous agents to build adaptive, intelligent systems. I am currently building web agents that operate directly in the browser: the LLM reads pages as compact text, acts through small, verifiable steps, and pauses to ask before taking any irreversible action. I also study how to make LLMs steerable through prompt optimization across tasks and languages.
High-quality synthetic data creation to improve model training, robustness, and performance. I work on generating and selecting data for training, evaluation, and preference learning, so models learn from the most informative examples.
Efficient training strategies with strong performance guarantees for deep networks and large-scale models, from efficient training and fine-tuning to fast inference for large models.
I am a Research Scientist at LG AI Research in Ann Arbor, Michigan, where I work on large language models end to end: the agents built on top of them, the data they learn from, and the optimization methods that train them efficiently at scale.
I received my Ph.D. in Electrical Engineering from Stanford University, advised by Prof. Mert Pilanci. My dissertation developed convex optimization foundations for training neural networks, work recognized with the Adobe Research Fellowship and an ICASSP Outstanding Paper Award. That background still shapes how I approach LLMs today: with an eye for principled, efficient methods.
During my Ph.D., I interned at Google Research with Behnam Neyshabur and Harsh Mehta, and at Salesforce Research with Yu Bai. I received my B.S. and M.S. in Electrical and Electronics Engineering from Bilkent University, working with Prof. Suleyman Serdar Kozat.
NeurIPS 2021 · top 8% of reviewers
ICASSP 2021
DeepMath 2020 · invited talk at DeepMath 2021
