Redline Ranch

Inspiration

Heart health is invisible until something goes wrong. We wanted to make Apple Watch cardiac data feel real and actionable not buried in the Health app. HenHacks' Wild West theme gave us the perfect universe to build it in.

What it does

Redline Ranch streams live ECG, heart rate, HRV, VO2 max, and SpO2 from your Apple Watch to a real-time dashboard. You can start a BlazePose-powered workout that counts reps automatically via webcam, compete with teammates in co-op mode with live-synced stats, and chat with Heartly an AI assistant that reads your actual ECG voltage data and explains your heart rhythm in plain English.

How we built it

Frontend: React-Native framework using Vite Backend: Python using FastAPI iOS App: We used Xcode to program the app, and the Healthkit API to get health data Chatbot: We used the Gemini API request Trained Model: We used a Functional CNN to classify ECGs from the Apple Watch

Challenges we ran into

Real-Time Syncing Ensuring heart metrics updated live without requiring manual refreshes required anchored queries, polling fallbacks, and careful state handling.

Pose Detection Stability BlazePose inference is computationally heavy. We implemented async loop controls, UI throttling, and smoothing to prevent freezes and rep overcounting.

ECG Signal Handling Raw ECG signals contain thousands of samples. We needed to balance performance with waveform interpretability.

Full-Stack Coordination We integrated SwiftUI, HealthKit, Python backend APIs, React, TensorFlow.js, and AI services all within hackathon time limits.

Accomplishments that we're proud of

Rep counting that actually works. Heartly analyzing real ECG voltage data — not just metadata. A full co-op multiplayer session that syncs two computers live with just a 6-character room code.

What we learned

Real-time syncing is complex. Working with HealthKit required anchored queries, fallback polling, and careful state management to ensure heart metrics updated reliably across devices. Browser-based ML demands performance discipline. Running TensorFlow.js BlazePose in real time forced us to optimize inference loops, smooth noisy joint angles, and prevent UI freezes or rep overcounting. ECG data must balance detail and usability. Raw ECG recordings contain thousands of voltage samples. We learned how to downsample intelligently while preserving waveform structure and interpretability. Health tech requires responsible design. When presenting cardiac data, clarity and proper disclaimers are essential. We designed our AI assistant to be educational, encouraging, and safety-conscious. This project strengthened our full-stack integration skills and showed us how powerful wearable data can be when transformed into actionable insight.