Emergency intelligence that works anywhere. No internet. No cloud. No accounts.
-3DDC84?style=for-the-badge&logo=android&logoColor=white){kind=icon}
ResQ is an on-device emergency intelligence companion built for the moments between an incident and professional medical care — when there is no internet, no doctor, and no one to tell you what to do. It uses Gemma 4 E2B — running entirely on your device — to analyse emergency situations through camera, voice, and text, then provides calm, structured guidance in five scannable cards.
🏆 Winner — Best Multimodal AI at the Build with Gemma: AI for Africa hackathon (FUTMinna 2026). See the winning writeup.
Zero network requests after model download. A live privacy counter proves it.
It is 11:30 PM at FUTMinna. A student walks back from night class at Gidan Kwano campus. A snake bites his ankle. The university clinic is closed. The nearest hospital is 4 hours away. His phone has no signal.
In rural Nigeria, this is everyday reality. The minutes between an incident and professional care are when permanent harm happens — tourniquets on snake bites, butter on burns, moving fracture victims incorrectly.
Existing solutions need internet, which is the first thing to go. ChatGPT cannot help when there is no signal.
| Feature | Description |
|---|---|
| Multimodal Emergency Analysis | Camera captures injury photos; voice records your description. Gemma 4's vision encoder and language model analyse both together. |
| Structured Guidance | Five calm, scannable cards: Assessment, Actions, Avoid, Monitor, Seek Care. Never raw AI text. |
| Medical Summaries | Professional reports formatted for healthcare providers — timeline, symptoms, findings, actions taken. |
| Complete Privacy | All processing is on-device. A live network monitor proves zero outbound requests. Encrypted medical profiles. |
| Offline Facilities | Local JSON database of medical facilities sorted by GPS proximity. Works without internet. |
| Care History | Every emergency session saved locally with full timeline, images, and generated reports. |
| Auto Model Download | Gemma 4 E2B (~2.4GB) downloads on first launch with progress tracking and comforting messages. |
| Layer | Technology |
|---|---|
| Framework | Flutter 3.44+ (Android + iOS) |
| On-Device Model | Gemma 4 E2B via flutter_gemma 1.3.0 + flutter_gemma_litertlm (LiteRT-LM engine) |
| State Management | Riverpod |
| Routing | GoRouter |
| Local Database | SQLite (sqflite) |
| Encryption | flutter_secure_storage |
| Typography | Bundled Inter font. No runtime fetches. |
| Icons | HugeIcons |
The design point: structured UI owns the experience; Gemma owns the understanding.
flowchart TD
A["📸 Camera + 🎤 Voice\n(multimodal input)"] --> B["🧠 Gemma 4 E2B\n(on-device · LiteRT-LM)"]
B --> C["📋 Structured Guidance\n(5 Flashcards)"]
B --> D["📄 Medical Summary\n(Healthcare Report)"]
C --> E["🏥 Continue To Care\n(Facilities · Contacts · Share)"]
D --> E
E --> F["🗄️ Care History\n(SQLite · Encrypted)"]
F -.->|past sessions| B
style A fill:#1A1A2E,stroke:#1B5E5C,color:#1B5E5C
style B fill:#1A1510,stroke:#C62828,color:#C62828
style C fill:#1E2640,stroke:#2E7D32,color:#F3EFE6
style D fill:#1E2640,stroke:#1B5E5C,color:#F3EFE6
style E fill:#1E2640,stroke:#4A6361,color:#F3EFE6
style F fill:#1E2640,stroke:#232C44,color:#8A8F9C
- Deterministic code owns: guidance card rendering, medical summary formatting, GPS proximity calculations, encrypted profile storage, local facilities database.
- Gemma does what only a model can: analyse images and voice together, determine emergency type and severity, generate structured JSON guidance, write professional medical summaries.
- Flutter 3.44+
- Android device with 4GB+ RAM (for on-device inference)
- iOS 16.0+ device
- A free Hugging Face read token
The app downloads the model automatically on first launch — no manual setup needed. It fetches Gemma 4 E2B (~2.4GB) from Hugging Face with a progress bar and comforting messages.
Manual setup (optional — for developers)
Download from litert-community/gemma-4-E2B-it-litert-lm (ungated, Apache 2.0):
# Android
adb push gemma-4-E2B-it.litertlm /sdcard/Android/data/com.resq.resq/files/
# iOS
# Use Finder/iTunes File Sharing to copy the model into the app's documentsgit clone https://github.com/THEJOHNCALEB/resq.git
cd resq
cp config.example.json config.json
# Edit config.json with your Hugging Face token
flutter pub get
flutter run --dart-define-from-file=config.jsonflutter test6 tests verify the deterministic pipeline: emergency session serialization, medical profile encryption, JSON roundtrip integrity — no model required.
lib/
├── core/ # Constants, routing, theme, privacy monitor
├── features/ # One folder per feature (emergency, home, profile, ...)
│ ├── <feature>/
│ │ ├── data/ # Models, repositories
│ │ └── presentation/
│ │ ├── pages/ # Screens
│ │ └── widgets/
├── shared/
│ ├── providers/ # Riverpod providers
│ ├── services/ # GemmaService, agent, facilities, location, database
│ └── widgets/ # Reusable widgets
assets/
├── data/ # Offline facilities database
├── fonts/ # Bundled Inter font
└── images/
Contributions are welcome! Please read CONTRIBUTING.md first — it covers development setup, branch conventions, and the pull request process. For security issues, see SECURITY.md.
| No data leaves the device | Every image, voice recording, and medical profile is processed on-device. The privacy seal is a live network counter, not a badge. |
| No runtime font fetches | Inter is bundled as an asset — google_fonts was removed entirely to keep the "0 requests" seal literally true. |
| No real data in the repo | .gitignore blocks *.task, *.bin, *.litertlm, *.m4a, config.json. |
| Auditable proof | HttpOverrides + fetch patching intercept and count every outbound request across all platforms. |
Pre-built APKs are available on the Releases page.
The CI/CD pipeline (.github/workflows/release.yml) automatically builds release APKs on tagged pushes:
git tag v1.0.0
git push origin v1.0.0- Google Gemma — the on-device model powering all AI features (Gemma is a trademark of Google LLC)
- flutter_gemma — the Flutter plugin for on-device Gemma inference
- Build with Gemma: AI for Africa (FUTMinna 2026) — the hackathon where ResQ won Best Multimodal AI
Built for Build with Gemma: AI for Africa — FUTMinna 2026
Gemma is a trademark of Google LLC.