1. Intro
TL;DR: I wanted to see what would happen if I gave an LLM direct control of my oscilloscope to help me decipher waveforms from a Class D amplifier in my kids’ audiobook player, and ended up with a general tool that’s helped me blind-characterize two power supplies from ripple alone, decode an unknown UART stream from raw analog samples, and find broken firmware commands. Code and raw data are here.
Disclaimer: I slept through two years of electronics in my CS degree, and my practical skills topped out at Ohm’s law. But I promised myself that if I ever took a break from work, I’d have a stab at electronics again. This has been mostly a way to improve my basic signal analysis techniques, and fix a few broken toys around the house.

Figure 1: A USB PD charger’s output ripple, captured and identified by an AI driving the scope over the network. The textbook PFM sawtooth — slow capacitor discharge, then a burst of switching pulses — was found without any hints about the hardware.
Continue reading I automated my scope to decipher wave forms and instead it found firmware bugs.
