Gearbox HITL Data Validator

A greenhouse ML team needed to label thousands of tomato images to train their ripeness model. No tool existed. I built one.

Annotators navigate the image grid with keyboard shortcuts (WASD / arrow keys / hjkl), assign a ripeness stage with a single keypress (1–9), and inspect any image full-size with the spacebar. Rated images fade to greyscale so focus stays on what still needs work. No mouse required.

Demo uses randomised sample data — the original client backend is not included.