As we develop artificial intelligence with capabilities far exceeding our own, the most profound challenge is not simply scaling their power, but encoding our values and character. The ultimate measure of these systems will be not their raw intelligence, but their resonance with our own humanity. Without this resonance, an AI might follow a command with a catastrophic literalness that misses the point entirely, or pursue a metric with a ruthless efficiency that violates our deepest-held norms. My research focuses on this critical process of guidance: learning about the nuanced fabric of human factors to build beneficial partners, and learning from our unique ingenuity to build more capable machines that can co-create with humans seamlessly.

At Google DeepMind, my work involves building some of the largest and most capable models today. I was a core contributor to the training of the Gemini family of models (1.0, 1.5, and 2.5) and also helped develop the open-source Gemma models.

My passion for collaborative AI began during my PhD, where my research explored steerability, reliability, and complex problems at the intersection of LLMs and RL. My work on reinforced inference-time alignment was honored with a Best Paper Award at AAAI 2021.

You can find all my publications on my Google Scholar.

I'm always open to interesting conversations. Feel free to reach me at my personal email or work email.