Hi, I'm Rob Porter.
I'm a physicist by training — my PhD was in experimental particle physics, searching for new physics in collider data. Since then I've spent my career building machine learning systems: during grad school I founded GradeCam, where we used computer vision to let teachers grade paper assignments with any webcam, and more recently I've led AI at PicnicHealth, teaching language models to read medical records.
This blog is where my personal experiments live. The current one started with a question I couldn't let go of — why are transformers so bad at math? — and ended, several detours later, with DOOM's renderer compiled into the weights of a stock transformer. No training anywhere. The story starts with the compiler I built to find out.
When I'm not doing that, I've converted a 1972 MGB to electric drive and reproduced a 1975 S-100 video board as kits for the retrocomputing crowd.