We went through the exciting journey of optimizing Neural Amp Modeler DSP to run inference real-time on the tiny Daisy Seed Arm Cortex-M7. Think hand-rolled GEMM kernels for small matrices and other fun stuff :D