I let an LLM write a custom kernel for doing inference on an RPi cluster. The age of bespoke custom kernels is upon us.