About

Small interactive examples of the ideas behind my work. Choose one to explore.

TurboQuant Vector quantization Latent reasoning Exploration for RL Math olympiads Data & evaluation Graph search Nearest neighbors Data sketching Coordinated sampling KDEformer Efficient attention

I’m a PhD student in Computer Science at New York University, advised by Christopher Musco and affiliated with Theoretical Computer Science at NYU. I’m also a 2025 Apple Scholar in AI/ML.

I develop algorithms for more capable and efficient language models. My work spans pretraining, post-training with reinforcement learning, and inference, as well as the data used to train and evaluate models. I study how models retain and use long contexts, how exploration improves their reasoning, and how to evaluate the correctness of that reasoning.

I combine randomized algorithms and mathematical analysis with empirical work on model architectures, training, and inference. The goal is to turn an understanding of how models work into measurable gains in accuracy and efficiency. I co-developed TurboQuant, a vector quantization algorithm for language model inference and vector search with near-optimal distortion guarantees.

Before my PhD, I earned a B.S. at the University of Tehran, where I worked with Prof. Goharshady and Dr. Zandieh.