Automated Fruit Quality Inspection & Sorting

Sorting fruit by quality is normally a manual, eyeballed process — slow, inconsistent between inspectors, and hard to scale. The capstone brief was to replace that judgment call with a system that could see a piece of fruit, decide on its quality, and physically sort it, without a person in the loop.

An ESP32-CAM streams live video to a React web interface, which sends frames to the Google Gemini API for quality analysis. The model's JSON response drives sorting commands sent over the local network to a six-DOF robotic arm and a NEMA 23 stepper-driven turntable, which route each piece of fruit to the right bin. A custom power supply was designed from scratch to run every subsystem — camera, arm, motors, and controller — off one board.

A working end-to-end prototype: camera in, AI judgment, robotic action out, with no manual step in between. It's the clearest demonstration of the same span this whole portfolio is about — circuit design, embedded firmware, a web interface, and an AI model, all built by one person to solve one physical problem.