AI-Powered Nursing Handover Automation – Flutter Mobile App

MediRelay is a Flutter-based mobile solution that automates the creation of nursing handover documents. It extracts and summarizes key information from EMR nursing records and generates standardized handover reports.

flowchart LR
    A[Flutter App] -->|Audio Upload| B[Whisper API Server]
    B -->|Transcript JSON| C[Backend API Gateway]
    C --> D[AI Processing Layer]
    D -->|Summary + Slots| E[Handover Generator]
    E -->|Standardized Report| A
    A -->|QnA Request| D

• Frontend: Flutter 3.16 (Dart 3.0)

• Backend Gateway: Node.js (Express, Passport.js, JWT, RBAC)

• AI Processing Layer:

  • Whisper API (Speech-to-Text)
  • KoBERT (Slot classification)
  • KoBART (Summarization)
  • OpenAI GPT API (Fallback summarization/QnA)

• Database: PostgreSQL (server), SQLite (mobile local cache)

• Networking: REST API (JSON), dio (Flutter client)

POST   /api/auth/login
POST   /api/auth/logout

GET    /api/patients               # List of patients
GET    /api/patients/:id           # Patient details

POST   /api/voice/upload           # Upload audio file (m4a/webm)
→ Calls Whisper API → Returns transcript

POST   /api/records                # Save nursing record
GET    /api/records/:patientId     # Retrieve patient’s records

POST   /api/handover/generate      # Generate handover document
GET    /api/handover/:patientId    # Retrieve handover document

POST   /api/chat/query             # Submit query

• Input: .m4a / .webm audio file
• Output: JSON { "transcript": "..." }

• Input: Transcript text
• Output: { "vital_change": true, "medication": "Tylenol", ... }

• Input: Nursing text + Slot results
• Output: One-line summary (e.g., "BT 37.9℃ → Tylenol administered, monitoring fever response")

• Input: Slots + summary results
• Output: Standardized handover document (JSON/Markdown)

lib/
├── core/               # Constants, networking, utilities
├── features/
│   ├── patients/       # Patient list
│   ├── voice_recording # Audio recording + Whisper API integration
│   ├── nursing_records # Record creation and retrieval
│   ├── handover/       # Automated handover generation
│   └── chat/           # QnA chatbot
├── shared/             # Common widgets
└── app.dart

1. Patient Selection

   • /patients/:id → Displayed in Flutter app

2. Voice Recording

   • Flutter → /voice/upload → Whisper API → Transcript JSON

3. Slot Classification & Summarization

   • Transcript → /handover/generate
   • KoBERT classification + KoBART summarization → Draft handover

4. View Handover

   • Flutter app fetches patient’s handover document

5. QnA

   • Flutter → /chat/query → GPT API → Response

cd backend
npm install
npm run dev

flutter pub get
flutter run

• Built with React + Vite (QR scanning, voice recording → Whisper API, record/handover/chatbot validation)
• Deployed on GitHub Pages, tested user flows and AI summarization

• Rebuilt as a mobile-first app
• Features: voice recording, Whisper API integration, OpenAI API-based summarization and handover generation
• Introduced structured state management (Bloc), local caching (SQLite), and backend persistence (PostgreSQL)

• 2025 Seoul AI Hub – SAY Track Finalist & Graduation
• 2025 Yonsei University Y-Compass Student Startup Team (Grade A completion)
• 2025 Pusan National University – 9th Digital Healthcare MEDICAL HACK, Grand Prize (President’s Award)
• 2025 Ministry of Education – Top 300 Student Startup Teams (U300+) Finalist

• Web Prototype Repository
• Service Intro Page (Notion)