MediVoice AI

Speak your symptoms, get instant AI-powered health guidance — making quality healthcare accessible to everyone using Amazon Nova.

Comment

Inspiration

Millions of people worldwide cannot afford doctors or live far from hospitals. Every day, people suffer because they don't know if their symptoms are serious or not. A mother in a rural village cannot get medical guidance for her sick child. An elderly person living alone doesn't know if their chest pain needs emergency care. This reality inspired MediVoice AI — what if anyone, anywhere could simply speak their symptoms and get instant, reliable health guidance powered by Amazon Nova AI?

What it does

MediVoice AI is a voice-powered medical assistant that allows users to:

  • 🎤 Speak or type their symptoms in natural language
  • 📸 Upload a photo of a visible condition like a rash or wound
  • 🧠 Receive instant AI analysis with possible conditions and next steps
  • 🔊 Get voice responses spoken back using Nova Sonic
  • 🌍 Work on any device — phone, tablet, or computer
  • ✅ Always shows medical disclaimer for responsible use

How we built it

The project uses three Amazon Nova models working together:

Model Role
Nova Sonic Converts voice input to text and speaks responses
Nova Lite Analyzes symptoms and reasons about conditions
Nova Multimodal Embeddings Processes and analyzes uploaded medical images

Tech Stack:

  • Backend: Java Spring Boot with LangChain4j
  • Frontend: React.js with Web Speech API
  • AI: Amazon Nova via Amazon Bedrock
  • Storage: AWS S3 for secure image storage
  • Build: Maven

Challenges we ran into

  • Nova Sonic bidirectional streaming required HTTP/2 configuration with NettyNioAsyncHttpClient which was complex to set up in Java
  • AWS SDK integration with Spring Boot needed careful configuration of Bedrock runtime clients
  • Voice recording in browser required handling microphone permissions and audio encoding correctly
  • Timeout handling for Nova Sonic TTS required CompletableFuture with time caps and graceful fallback
  • CORS configuration between Java backend and React frontend needed careful setup

Accomplishments that we're proud of

  • ✅ Successfully integrated all 3 Amazon Nova models in one application
  • ✅ Built a working medical analysis system that responds in ~13 seconds
  • ✅ Implemented secure image storage in AWS S3 with unique file naming
  • ✅ Created a clean, accessible UI that works for elderly users
  • ✅ Added responsible AI features with medical disclaimers throughout
  • ✅ Built graceful fallbacks when Nova Sonic is unavailable

What we learned

  • How to orchestrate multiple Nova models in a single Java pipeline
  • The power of multimodal AI for combining voice, text and image inputs
  • How LangChain4j simplifies AI agent development in Java
  • The importance of responsible AI design in healthcare applications
  • How Amazon Bedrock makes it easy to access frontier AI models

What's next for MediVoice AI

  • 🌐 Add multilingual support using Nova Sonic's language capabilities
  • 📱 Build a mobile app for offline rural use
  • 🏥 Integrate with real medical databases like WHO and NHS guidelines
  • 🤝 Partner with NGOs and community health workers
  • 🔒 Add patient privacy features and HIPAA compliance
  • 📊 Build symptom tracking over time for chronic conditions

Built With

Share this project:

Updates