I'm a final year PhD student at MIT CSAIL working with David Sontag. I'm interested in self-improving agentic systems that can learn effectively from weakly supervised data and limited expert oversight, often in healthcare contexts.
Most recently, I was an intern at Layer Health, training LLM agents for clinical decision guidance and designing evals. Previously, I interned at Apple, developing methods and evals for multimodal reasoning with LLMs,1 and at Microsoft, learning representations of the adaptive immune system.2
My long-term goal is to build AI systems that make humans smarter. I picture such systems as supple instruments that supercharge the transition from data to information, collapsing the distance between questions and answers. I believe the biggest impact of AI will not be what it can do in our stead, but the pace at which it enables us to grasp the world and act upon it. From that acceleration, I think, will come the next great enlargement of human capability.
research
My PhD research has revolved around the following themes:
LLM agents + representation learning + sample-efficiency: How can LLM agents learn and design good representations of data that enable sample-efficient downstream learning?13
Causal inference + sample-efficiency: How can we integrate large-scale real-world evidence with small-scale experimental data to power valid, unbiased, and efficient causal inference?45678
I have also worked on AI for science, such as learning representations for the adaptive immune system2 and predictive modeling with omics data.910 Before my PhD, I worked on reinforcement learning for type-1 diabetes management11 and dabbled in communication networks theory.1213
selected work
see all publications or google scholar for the complete list.