Engineering notes
Short, technical write-ups on the internals — one or two a week.
An interactive look at anisotropic descent
A hand-coded TypeScript figure: gradient descent on a stretched quadratic bowl, with a live learning-rate control. Plus the deploy recording for this very site.
visualizationtypescriptoptimizationRecording the tape as a flat array
The data structure under adjoint AD is usually a graph of heap nodes. We use a struct-of-arrays trace instead, and the JVM rewards it with predictable cache behaviour.
internalsautodiffjvmWhy NablaTensor exists
Monte-Carlo books recompute the same valuation millions of times. We record it once as a flat trace and replay it — one forward sweep for the price, one reverse sweep for every Greek.
metaautodiff