Skip to main content
Caleb Hunter
Hackathon Submission

HealthBridge

Real-time ASL-to-text communication bridge for deaf patients and doctors

466 million people worldwide have disabling hearing loss. HealthBridge is a real-time communication bridge using Gemini 3 Flash for ASL recognition and bidirectional translation — built in 48 hours at a hackathon.

AI Involvement

Gemini 3 Flash is the core engine of HealthBridge. It processes live video frames from the WebRTC stream, recognizing ASL gestures in real time and translating them to text for the doctor. The same model handles the reverse — converting the doctor's speech responses into structured text output for the patient. We deliberately chose Gemini 3 Flash over larger models because latency was the primary constraint: a communication tool in a clinical context needs sub-second response times, not the deepest reasoning. The AI isn't a feature — it's the entire bridge.

Key Technical Decisions

  • Gemini 3 Flash over larger models — speed was the primary constraint for real-time ASL recognition
  • LiveKit for WebRTC infrastructure — managed media server handles the low-latency video pipeline
  • Bidirectional translation architecture — ASL→text for patient input, speech→text for doctor responses
  • Stateless session design — no persistent storage of health conversations, privacy by default
  • Built in 48 hours — scoped to a working demo rather than production hardening

Tech Stack

Next.jsTypeScriptGemini 3 FlashLiveKitWebRTCVercel