Everything you need to start building.
Tools, data and guides picked for health projects, plus our ground rules for building with health data.
Tools and guides
AI tools
- Claude Models for building and coding
- Hugging Face Open models and datasets
- Teachable Machine Train a model with no code
- MediaPipe Vision models that run on a phone
- Whisper Open speech-to-text
Health data
- PhysioNet ECG, EEG and sleep recordings
- Kaggle datasets Thousands of public datasets
- CDC open data U.S. public health statistics
- openFDA Medicines and safety reports
- Synthea Synthetic patient records
Learn the basics
- Hack Club Workshops Beginner-friendly project guides
- MDN Web Docs How the web works
- Kaggle Learn Short, free machine learning lessons
- MedlinePlus Plain-language health topics
Data to start with, by part of the body
Brain
- Sleep-EDF Overnight sleep recordings
- CHB-MIT EEG recordings of seizures
- 988 Lifeline Where crisis features should point
Eyes & Ears
- MediaPipe Hand, face and pose tracking
- Whisper Open speech-to-text
- APTOS retina images Diabetic eye disease
- VizWiz Photos taken by blind users
Lungs
- AirNow API Official U.S. air quality
- OpenAQ Open air-quality data worldwide
- CDC open data Asthma and health statistics
Heart
- PTB-XL 21,000+ public ECGs
- MIT-BIH Arrhythmia Classic heart rhythm data
- MediaPipe Pose Body tracking for CPR form
Gut
- Open Food Facts Millions of food labels
- USDA FoodData Central Nutrition for every food
- Food-101 101,000 labeled food photos
Legs
- MediaPipe Pose 33-point body tracking
- MoveNet Fast pose tracking in the browser
- Kaggle datasets Sports and fitness data
Whole body
- openFDA Drug labels and safety reports
- MedlinePlus Plain-language health info
- Synthea Realistic synthetic patients
Build it responsibly
Health tools can help or hurt. These rules keep your project on the right side, and judges reward them.
- Use public or synthetic data
- Never real patient records, and never anyone’s private health information, including your own.
- Don’t play doctor
- Give information, not diagnoses, and always say when someone should see a professional.
- Fail safe
- When your tool is unsure, it should say so. In an emergency, it should point to 911 or 988 first.
- Collect less
- Only ask for the data you need, and explain why you need it.
- Test across people
- Different skin tones, accents, ages and abilities. AI that only works for some people isn’t done.
- Be honest about limits
- Tell users what your project can’t do. Judges reward it.
Workshops we’re planning
Free sessions in spring and at the event. Topics may change.
- Your first AI app in an hour
- Working with health data
- Computer vision with your webcam
- Designing for accessibility
- Building responsibly with health AI
- Demos that land