Arya Somu

I was born in San Francisco, CA, live in Seattle, WA, and go to school at UC San Diego.

I co-founded Laplace Research, an AI research lab building the future of financial world models. I also operate Monolith Systematic, a macro investment manager.

Right now, I'm working as a research assistant at Stanford's AFT Lab, where I'm working on rare-event algorithms. I'm also an Engineer at Alexandria, a startup focused on agentic orchestration.

Recently, I've worked with great people and teams at Esplanade Capital, AfterQuery, Aletheia Laboratories, and Arizona State's Beus Center.

You can find me on X, GitHub, LinkedIn, Curius, Scholar, and Substack, or reach me via email.

Notable Work

  • Yolk

    A Modern E-Graph Engine for OCaml.

  • Sparse Parity Learning

    An MLP hits a real sample-complexity wall learning sparse parities, with a Fourier-basis progress measure that sees it coming before the loss does.

  • Grokking on Modular Arithmetic

    Replicating grokking (delayed generalization) in a small transformer trained on modular arithmetic.

  • Anvil

    A CLI coding agent that knows when not to edit. Constrained, auditable changes instead of maximal autonomy.

  • Robinhood

    Prediction markets reprice events faster than equities. This engine trades the gap through Robinhood's agentic MCP.

  • Xenith

    Rust primitives for keeping state honest across chains, with divergence surfaced instead of silently resolved.

  • Vela

    The fastest published verifiable DEX matching engine ever.

    Paper: Vela Exchange
More Work