Arya Somu

Judgement and calibration are the bottlenecks to good decisions under uncertainty. Models can now produce forecasts at scale, but not ones whose confidence you can take at face value. Laplace Research is our attempt at solving the latter, so that human judgement can be spent choosing which questions matter rather than second-guessing the answers.

I study CS and Math at the University of California, San Diego, build enterprise intelligence at Alexandria, and research rare events at Stanford's AFT Lab. I'm broadly interested in continual learning, exchange operations, and the frontier of modern mid-training.

Previously, I interned on Esplanade Capital's fixed income desk, where I worked on prepayment modeling and OAS analysis. Before this, I did research on subatomic forces, managed a 500+ transaction resale business, and tried competing in competitive math, physics, and programming. When I'm free, I'm typically reading technical reports or forecasting literature, playing very competitive poker, lifting, or playing badminton.