When this build is running, a small color display rides along on a camera gimbal and shows how the rig is moving up and down. A white center line sits on screen with a green leveling bar next to it, and as the gimbal is tilted to the left and brought back upright, the green bar changes shape and angle against that line. A two-digit readout on the same screen shifts between values like 94, 92, 91, 90 and 96. The point of all this is the z-axis, the up and down motion that a gimbal leaves unstabilized, and which depends on how steadily the person holding it walks.
The hardware is compact: a Seeed Studio XIAO ESP32-C3 board, a BN0085 sensor, an ST7735 display, some buttons and cables, and a small LiPo battery rescued from a bike light. The whole set of parts came from AliExpress for under 25 dollars. The maker picked the ESP32-C3 because it is tiny, cheap and has exactly the GPIO pins needed. Everything lives in a grey PLA case with corner screws, a metal push button and a square button beneath it, attached to the gimbal through a bracket on its side.
The process behind it is the part that stands out. The maker wrote none of the code personally, leaning on AI for the firmware, the case design drawn up in Fusion over an MCP connection, and even most decisions like part choice. The libraries are the Adafruit BN0085, ST7735 and GFX ones. The reported result is that the AI-written code worked perfectly.
There are open ends worth noting. I2C refused to cooperate, so the switch went to SPI at the cost of more pins, and routing cables inside the cramped enclosure was a fight. The displayed line is a moving average rather than the exact sensor readout, and how steady that reads during a real walk still seems worth testing.