Sample GenAI Plans to Generate Fast FFT Algorithms
Assembly FFT Plan 1
Build a 512-point radix-2 FFT with the arithmetic hand-written in ARM assembly, callable from MicroPython on stock firmware. Verified against 10 synthetic audio signals with cycle-accurate timing. Complete — 885 µs per FFT, 178× faster than pure MicroPython, 10/10 signals pass.
Assembly FFT Plan 2
Build a family of competing implementations — real-input FFT, radix-4, specialized stages, dual-core, hand-encoded instructions — measured against each other under one harness, so students can discover the architecture tradeoffs by measurement rather than assertion. Executed — 8 variants built and measured; best is 1.41× the baseline.
Hands-On Lab Series Plan 3
Design 32 hands-on labs taking a student from "never opened Thonny" to a hand-written assembly FFT they have benchmarked and can explain. Assumes no prior FFT, DSP, or assembly experience. Includes the learning-graph expansion (200 → ~550 concepts) and course-description updates. Draft for review.