[amgad]

GrowCare 2

Built at INTI International University, 2026

Fully offline agriculture assistant for Android. A YOLO11 model detects plant disease on the phone across 38 crop categories, and Gemma through LiteRT answers farming questions with no internet connection.

ROLE
"Team leader"
PERIOD
"2026"
CONTEXT
"AI course, INTI International University"
STACK
"Kotlin, YOLO11, Gemma, LiteRT, Room"
UPDATED
"Sep 2026"
github.com/amga-d/GrowCare2

Overview

GrowCare 2 turns a phone camera into a crop diagnostic tool that works with no connection at all. It is built for farmers in rural areas where the network can't be relied on.

A YOLO11 classifier identifies plant disease across 38 crop categories on the phone. Its result is injected into the prompt of a Gemma model running locally through LiteRT, so the text-only model answers as if it had seen the photo. Scans and chat history stay on the device in a Room database.

What I built

  • On-device vision: YOLO11n-cls exported to TFLite FP16, with EXIF handling so camera orientation doesn't break inference.
  • On-device chat: Gemma through LiteRT-LM, with up to 10 turns of conversation memory.
  • A hybrid pipeline that feeds the classifier's result into the language model's prompt.
  • Kotlin and Jetpack Compose interface, with Room for local persistence.

How it fits together

    • Cameraphoto of a leaf
    • YOLO11n-clsLiteRT, 38 classes
  1. label into prompt
    • GemmaLiteRT-LM
    • Roomscans, chat history

Every step runs on the phone. There are no network calls.

In the repository

$ git clone https://github.com/amga-d/GrowCare2.git

tree -L 1

.
├── .github/
├── .idea/
├── app/
├── gradle/
├── ML/
├── .gitignore
├── AI_SYSTEM_ARCHITECTURE.md
├── build.gradle.kts
├── gemma.md
├── gradle.properties
├── gradlew
├── gradlew.bat
├── README.md
└── settings.gradle.kts

git log --author=amga-d --oneline -5

  1. 6dc2151Merge pull request #1 from amga-d/develop
  2. bfad4eerefactor: remove retired product documentation and architectural plan files
  3. 86cbecadocs: finalize team name removal and YOLO instructions
  4. 1ccc163docs: update team info and add YOLO model conversion instructions
  5. 2a794eadocs: add setup and ADB push instructions to README
  • Kotlin 97.9%
  • Python 2.1%