Inspiration

🚑 LIFELINE — Project Story

Inspiration: Sudden cardiac arrest kills someone every 90 seconds, brain damage begins in ~4 minutes, but the ambulance takes 7–14 — so we built a paramedic-in-your-pocket for the panicked bystander in that gap, since survival drops ~10% per minute ($P_{\text{survival}} \propto e^{-\lambda t}, \lambda \approx 0.1$).

What we learned: AI in life-critical apps must be bounded, not creative — Claude Sonnet 4.5 never invents advice, it only routes messy human speech to one of 13 hard-coded Red Cross / AHA protocols.

How we built it: An Expo React Native frontend with a FastAPI backend running the voice loop (record → Whisper STT → Claude routing → gpt-4o-mini-tts steerable TTS) in ~2.4 seconds, with mood-based voice instructions (coach / urgent / calm), a 110-BPM CPR metronome, real OSM Overpass AED data, and offline fallbacks on every layer.

How Copilot helped: We used the GitHub Copilot REST API to auto-generate the 13 protocol JSON files (6 hours → 20 minutes), and Copilot inline wrote the entire animated ECG heartbeat SVG plus caught a React Native Web SVG-hydration bug before we shipped.

Challenges: Handling chaotic emergency audio through Whisper, parsing Claude's occasionally malformed JSON, forcing iOS audio playback to max volume, dodging OSM Overpass timeouts, and resisting every design instinct to add clutter — because if a bystander can't read it while doing CPR, it doesn't belong on the screen.

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for StackPhantom

Built With

• copilot
• openai