ChickDetect
A mobile app that screens chicken droppings for early signs of disease. It runs a machine-learning model fully offline on the phone and tells farmers what to do next.

Overview
Our capstone project, where I'm lead programmer. Farmers photograph a dropping, and the app classifies it on the device as healthy, coccidiosis, salmonella or Newcastle disease, then gives plain-language advice and vet contacts.
Problem
Small poultry farmers often notice disease too late, after it has spread through the flock. Lab tests and vet visits are slow and costly, and farms often have weak internet.
Solution
An offline-first Flutter app with an on-device TensorFlow Lite model. Droppings are one of the earliest visible signs of illness, so a quick photo becomes an early warning, with clear next steps instead of raw percentages.
Features
- Detects healthy, coccidiosis, salmonella and Newcastle disease from a photo
- Rejects photos that aren't droppings, plus photo-quality checks and a photo guide
- Runs fully offline with an on-device TensorFlow Lite model
- Scan history with a flock health summary and rescan reminders
- Step-by-step care advice and vet contacts for each result
- Download website with a per-device install guide
Technology Stack
Development Process
- 1Built the original Flutter app: camera capture, scan flow and history
- 2Integrated the disease-classification model on-device with TensorFlow Lite
- 3Added Filipino and English language support
- 4Team v2: UI/UX redesign, retrained model and release build
- 5Published a download website for farmers
Challenges
- Keeping the model accurate on real farm photos with uneven lighting and cluttered ground
- Avoiding false 'healthy' results on sick droppings, the costliest mistake
- Running inference fast and fully offline on low-end Android phones
Result
The v2 model reaches 96.56% test accuracy across 5 classes. A team capstone: I led programming and built the original app, and the v2 redesign and website were done with my groupmate.


