SafeDose

A verification and calculation tool for administering injectable medications prescribed by your healthcare provider.
Created by SafeLab, SafeDose helps you verify medication materials, perform unit conversion calculations, and reduce math errors when executing your provider's prescription.

Whether you're working with mg/ml, reconstituted peptides, or insulin units, SafeDose helps you verify your materials match your prescription and calculate the volume to draw.

---

🔮 VisionOS Vision

SafeDose today works in two steps:

1. Specify Dose: The user enters their prescribed dose (e.g., "2.5 mg Semaglutide")
2. Scan & Verify: They scan their supplies. The app uses computer vision to identify the syringe and vial, which the user confirms. It then calculates the volume to draw based on the verified information.

This mobile design is deliberate. It ensures SafeDose knows the intended dose before analyzing supplies, and it uses a guided verification flow with user confirmation at each step.

The longer-term vision is a glasses-native workflow:

1. Specify Dose: The user enters their prescribed drug and dose (e.g., "2.5 mg Semaglutide")

2. Automatic Recognition & User Verification: SafeDose recognizes the syringe type and vial in the user's field of view, which the user confirms

3. Guided Drawing with Verification: As the plunger moves, overlays show where to stop based on the user's prescription. The user confirms when correct.

4. Confirmation & Logging: The user confirms the dose matches their prescription, with optional logging for tracking

The current mobile flow is scaffolding. The end-state is an eyes-up, real-time verification system that helps users confirm their materials and execute their provider's prescription accurately.

---

⚠️ Disclaimer

SafeDose is a verification and calculation tool, NOT a medical device.

SafeDose helps you administer medications prescribed by your healthcare provider. We do NOT provide medical advice, diagnose conditions, or recommend dosages. This tool performs unit conversion calculations on information YOU verify and should not be used as a substitute for professional medical judgment.

Always follow your healthcare provider's prescription exactly. Confirm all calculations and materials match your prescription before proceeding. All results must be independently verified.

Use at your own risk. SafeLab and contributors assume no responsibility for misuse.

---

🚀 Hosted Version

A hosted version of SafeDose: https://app.safedoseai.com/

---

💻 Local Development

To run the open-source version locally:

1. Clone the repo

git clone https://github.com/your-org/rodneyg-safedose.git
cd rodneyg-safedose

2. Install dependencies

npm install

3. iOS (Xcode required)

npx pod-install
npx expo run:ios

4. Android (Android Studio required)

npx expo run:android

5. Web (experimental)

npx expo start --web

Note: To use the AI-powered scan feature, you’ll need your own OpenAI API Key.

⸻

🔧 Environment Setup

Create a .env file in the root:

OPENAI_API_KEY=your-openai-key-here



⸻

📁 Directory Structure

rodneyg-safedose/
├── app/                 # Screens and navigation
├── hooks/               # Camera + system hooks
├── assets/              # Icons and images
├── android/             # Android project files
├── ios/                 # iOS project files
├── docs/                # Documentation
│   ├── auth-config.md   # Authentication configuration
│   ├── MODEL_FINE_TUNING_GUIDE.md  # Complete fine-tuning guide
│   └── FINE_TUNING_QUICK_START.md  # Quick start guide
├── evals/               # Evaluation framework
│   ├── test-sets/       # Evaluation test cases
│   ├── exported-data/   # Captured user interaction data
│   └── README.md        # Evaluation documentation
├── types/               # Global type definitions
├── Evals.md             # Evaluation framework overview
├── package.json         # Dependencies
├── app.config.js        # Expo app configuration
├── tsconfig.json        # TypeScript settings



⸻

📊 Model Improvement & Evaluation

SafeDose automatically captures user interactions to continuously improve AI accuracy:

- **Automatic Data Capture**: Every dose calculation and AI scan is saved locally
- **Quality Scoring**: Automatic filtering ensures only high-quality examples for training
- **Easy Export**: One-click export to training-ready format for model fine-tuning
- **Privacy-First**: Images stored locally only, no personal data transmitted

**Resources**:
- **Quick Start**: [docs/FINE_TUNING_QUICK_START.md](docs/FINE_TUNING_QUICK_START.md) - 5-minute guide
- **Complete Guide**: [docs/MODEL_FINE_TUNING_GUIDE.md](docs/MODEL_FINE_TUNING_GUIDE.md) - Full instructions
- **Evaluation Framework**: [Evals.md](Evals.md) - Testing and benchmarking

⸻

🤝 Contributing

We welcome contributions focused on safety, accessibility, and clarity.

Steps:
	•	Fork the repo
	•	Create a branch (git checkout -b your-feature)
	•	Commit your changes
	•	Submit a pull request

Please avoid introducing breaking changes or external dependencies unless necessary.

⸻

🛡 License

MIT License.
By using this project, you agree not to market it as a certified medical device without independent validation and regulatory approval.

⸻

Built by SafeLab — advancing open tools for human systems.