Le DUTs

Benchy is an AI agent that controls real hardware to test, diagnose, and fix firmware/hardware deployments automatically. It bridges the gap between software CI/CD and physical hardware — every firmware change gets tested on real silicon with real measurements. The agent (powered by Gemini 3.1 Pro) orchestrates a Raspberry Pi 5 connected to a Digilent Analog Discovery oscilloscope, an FNIRSI DPS-150 programmable power supply, and two ESP32-S3 microcontrollers. It can set voltages, capture waveforms, decode protocols (I2C/UART/CAN), analyze signal integrity, and diagnose failures (three tier arch root cause analysis)—entirely through natural language. A phone camera streams live video to the Pi (flutter app, via websockets), letting the agent visually inspect the bench via Gemini vision. The agent guides users through wiring, runs experiments, performs root cause analysis with structured EE metrics (FFT, rise time, threshold comparisons), and iterates until the problem is solved. Hardware engineers today debug with stuff like tribal knowledge and screenshots in Slack. Benchy turns bench evidence into structured, reproducible CI artifacts — kinda like Vercel Preview Deployments, but for hardware.