Hi, I’m Behzad (بهزاد).
I work on large language models, with a focus on the post-training stage: supervised fine-tuning, reinforcement learning, reasoning, and evaluation. I spend most of my time thinking about how we train models to reason, behave, and improve after pretraining.
My training is in machine learning, and my work has ranged from theory to experiments to production systems. I currently work at Google DeepMind in the GenAI unit. Before diving into GenAI, I was making sense of the noise at Twitter (Cortex) by building tweet representations, and before that conducted NLP research at Megagon.ai.
I’m originally from Iran, where I studied computer science at the University of Tehran, before earning my PhD in Computer Science from Boston University in 2016.
Tehran – Home
Photo by Mohammad Amirahmadi
When I’m not tweaking reward functions, you can find me running or studying music theory.