-
-
LifeBridgeAI
LifeBridge: Medical-Grade Communication, Powered by AWS Generative AI
The Inspiration — A Lifeline in Critical Moments
In a medical emergency, clear communication is the difference between life and death.
For the deaf, hard-of-hearing, and non-verbal communities, that gap can turn a crisis into a catastrophe.
Our goal was to build a lifeline—a tool that empowers anyone to articulate their needs to first responders or clinicians, regardless of language, ability, or network connectivity.
LifeBridge is our answer: an equitable, accessible, life-saving communication platform.
What Is LifeBridge?
LifeBridge is a production-ready, multi-modal communication platform engineered for the extreme demands of healthcare. Acting as a universal translator, it seamlessly converts:
- Sign Language → medically contextualized text and speech
- Spoken Language → real-time transcription across 40 + languages
- Text → speech, with selectable voices and languages
The system is delivered as a resilient Progressive Web App (PWA):
- Runs on any device—from a paramedic’s tablet to an ER workstation
- Includes an offline mode to ensure it never fails
How We Built It — Engineering for Trust and Impact
Core Technology & AI Implementation
| Component | Details |
|---|---|
| AI-Powered Translation & Reasoning | - Amazon Bedrock with highly-tuned, context-aware prompts (backend/src/services/bedrock.ts) for nuanced medical translations.- Amazon Q ( frontend/src/services/amazonQService.ts) for intelligent triage suggestions and contextual advice. |
| Advanced Sign Language Recognition | Custom processor (backend/src/handlers/signLanguageProcessor.ts) uses landmark analysis to:1. Identify gestures 2. Assign medical priority ( critical, high, medium)3. Trigger priority alerts (e.g., validated emergency fist formation). |
| Medical-Grade Backend | Defined as Infrastructure-as-Code (backend/serverless.yml) on a HIPAA-aligned serverless stack:- Compute: AWS Lambda (Node 20.x) - Data: DynamoDB with encryption-at-rest (KMS), PITR, 7-year TTL audit logs - Real-time: AWS WebSocket APIs for low-latency, bidirectional comms |
A Multi-Layered Approach to Security & Compliance
- Automated PHI Redaction
backend/src/services/phiRedaction.tsautomatically scrubs 18 + categories of PHI before data hits any AI endpoint. - Immutable Audit Trails
backend/src/services/auditLogger.tsencrypts every interaction with KMS before storage. - Human-in-the-Loop QA
backend/src/services/qualityAssurance.tsflags low-confidence output, opens a review ticket, and uses Amazon SNS to notify licensed medical reviewers.
Design — Resilience & Usability Under Pressure
| Feature | Description |
|---|---|
| Offline-First PWA | Service-worker-powered caching of emergency phrases and a local-storage-backed offlineQueue (frontend/src/utils/offlineQueue.ts) so no data is lost during outages. |
| Dynamic Performance Modes | Toggle between: • Optimized — sub-second responses for emergencies • Standard — battery-saver for routine consults ( frontend/src/App.tsx) |
Challenges We Faced
- Latency vs. Medical Accuracy — Achieved sub-second responses without sacrificing terminology by optimizing Bedrock prompts and leaning on low-latency serverless design.
- Building Trustworthy AI — Combined PHI redaction, bias detection, and a robust human-in-the-loop review workflow.
- Unfailing Reliability — Architected a PWA + local queue to guarantee availability even with zero connectivity.
What We Learned
This project showed how cutting-edge AWS Generative AI can solve deeply human problems when trust, security, and resilience are foundational.
LifeBridge isn’t just an app—it’s proof that technology can build bridges, empower communities, and ultimately save lives.
Log in or sign up for Devpost to join the conversation.