Doctor-Green

Doctor-Green is a smart-farm platform that brings crop-disease diagnosis and IoT environment control onto a single web platform, so both live on one screen. I planned and built the whole pipeline solo, from sensor to dashboard.

It grew out of a VGG16 plant-disease-diagnosis app I built with team MAKENEW, which won the Excellence Award at the 2024 ICT·SW Women's Start-up Competition. That model never reached usable accuracy, so in 2026 I rebuilt the project as a full smart-farm platform and restarted the diagnosis model from the data up.

The IoT half runs today. ESP32 sensor nodes post environment data directly to Supabase over Wi-Fi, no relay server, and a Next.js web app visualizes it with 5-second polling. Actuators such as LEDs and fans are controlled through a desired-state pattern: the app only writes the target state, the ESP32 polls it and drives the hardware, and optimistic UI absorbs the round-trip delay. It is deployed on Vercel as a live demo.

The diagnosis model is still at the data stage. AI Hub blocks downloads from non-Korean IPs, so I wrote a Windows-side downloader that calls their API directly, merges the split archives, restores Korean filenames, and converts the JSON labels to YOLO-format boxes. From 5 disease classes × 1,000 images I extracted 39.9K lesion crops (train 32.2K / val 4,326 / test 3,408) using a group-aware stratified split, so crops from the same source photo never straddle train and validation. Training the crop classifier, ConvNeXt-Tiny or EfficientNetV2-S, is the next step; there are no results yet.