Writing about projects, ideas, and things I'm learning.
June 2, 2026
In ProgressI built a natural-language crypto wallet on Base, then spent most of the project trying to break it. The eleven transaction types were the quick part. The refusals were not.
June 2, 2026
In ProgressA local-first agent that decomposes prediction-market questions, argues against itself, and calibrates with gradient-boosted trees. The backtest numbers are suspiciously good, which is the most honest thing I can say about them.
June 2, 2026
In ProgressA news agent that reads eight politically diverse outlets, scores its own drafts on six axes of bias, and publishes an on-chain receipt proving the neutrality pipeline actually ran unmodified.
March 15, 2026
In ProgressI simulated 13,000 neurons from a real fruit fly nervous system, got emergent odor avoidance with no navigation logic, and now I'm putting it on a hexapod.
March 3, 2026
In ProgressMoving TAP-Score from a 2D pushing task to 7-DOF robotic arms. The simulator lied about everything.
March 3, 2026
In ProgressWe built a benchmark to test whether VLMs can reason about visibility, occlusion, and what's in frame. Most of them struggle more than you'd expect.
March 1, 2026
Testing a 3D U-Net dose prediction model against adversarial attacks and clinically plausible CT perturbations. Only one type of corruption actually breaks it. Now accepted for two conferences.
February 25, 2026
Building a scorer to catch bad robot actions, discovering that diffusion policy barely produces any, and what that means for the project going forward.
February 10, 2026
Welcome to my blog, a space for writing about projects, ideas, and things I'm learning.