Horizons

Inspiration

We were inspired by our parents who love yoga.

What it does

Horizons is a real-time wellness app that turns your webcam into a yoga coach. It detects your pose, checks your form, speaks corrections out loud, and pairs that with meditation flows, progress tracking, and a community wall.

How we built it

We built a React/Vite frontend on top of a Python backend that runs MediaPipe pose detection, pose classification, and form correction. Live video and session updates move over WebSockets, and Deepgram handles text-to-speech coaching.

Challenges we ran into

The hardest part was keeping coaching fast enough to feel live without wrecking accuracy. We also had to turn raw joint data into feedback people can actually follow, and deal with Reddit blocking our community scrape until we added a more reliable fallback.

Accomplishments that we're proud of

We got a full loop working in the browser: detect, classify, correct, and speak. Along the way we raised pose classification accuracy a lot, roughly doubled frame rate, and packaged it as a real product under the Horizons brand.

What we learned

A wellness coach only works if it feels responsive and clear, not just “smart.” We learned that latency, wording, and fallbacks matter as much as the model itself when you put computer vision in front of real users.

What's next for Horizons

Next we want smarter personalization from session history, better mobile support, and a stronger community layer. We’re also looking at broader pose coverage and safer beginner guidance.

Built With

• css3
• html5
• mediapipe
• python
• typescript
• yoga-82