I co-founded Eternis where we’re building systems for forecasting, reasoning and decision-making under uncertainty. In the past, we also worked on secure LLM inference and private AI deployments. Our recent work includes using RL to beat public state-of-the-art forecasting baselines and subsequently scaling a version of this recipe to much larger models, as well as using internal-representation probes to improve calibration, audit reasoning faithfulness, and reduce inference cost.

I completed my PhD at Johns Hopkins University under the wonderful supervision of both Matthew Green and Abhishek Jain. One of the key focuses of my dissertation work was the security and construction of efficient SNARKs and proof systems, particularly those involving error-correcting codes. I was a recipient of an EF academic grant that funded our work on SNARK security.