Automação de Notas Clínicas Compatível com HIPAA
sucheet2000/HIPAA-Compliant-Clinical-Note-Automation
Converte conversas médicas em prontuários FHIR-compliant com conformidade HIPAA
Ver no GitHubSistema de produção em Python que converte conversas clínicas em registros médicos estruturados e compatíveis com FHIR, com validação humana integrada. Implementa desidentificação de dados sensíveis (PHI masking), processamento via Claude AI e auditoria completa para conformidade HIPAA. Ideal para integração com EHR e automação de workflows clínicos em ambiente corporativo.
Instalação
npx convez-ai sucheet2000/HIPAA-Compliant-Clinical-Note-Automation1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automação de documentação clínica em consultórios e hospitais
- Integração com sistemas EHR existentes para reduzir carga administrativa
- Processamento de notas médicas com validação clínica em loop humano
- Conformidade regulatória e auditoria de transações médicas
Veredito

Projeto robusto com arquitetura profissional e conformidade regulatória explícita. Recomendado para equipes de saúde digital que precisam automatizar documentação clínica mantendo padrões enterprise e segurança de dados sensíveis.
README
A tradução para o português deste README ainda está na fila.
HIPAA-Compliant Clinical Note Automation Tool
A production-ready AI-powered healthcare system that converts clinical conversations into structured FHIR-compliant medical records with human-in-the-loop validation.
🎯 What This Project Does
Automatically extracts structured clinical data from unstructured doctor-patient conversations:
- De-identifies sensitive patient information (PHI masking)
- Processes conversations through Claude AI with structured outputs
- Transforms extracted data into FHIR R4 compliant medical records
- Validates through clinician dashboard with confidence scoring
- Audits all transactions for HIPAA compliance
Perfect for EHR integration, clinical documentation automation, and AI-powered healthcare workflows.
🛠️ Built With AI Tools
This project was developed rapidly using Claude AI and Google Gemini to demonstrate modern full-stack development practices:
- Claude AI: Architecture design, core algorithm implementation, structured output processing, API design
- Google Gemini: Code review, optimization suggestions, testing strategies
- Result: 8,500+ lines of production-ready code in one week
This showcases how modern developers leverage AI tools for rapid prototyping while maintaining code quality and architectural integrity.
⚡ Quick Start (3 Steps)
# 1. Setup environment
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
# 2. Start databases (requires Docker)
docker-compose up -d
# 3. Run the web interface
cd src && python main.py && python app.py
# Visit http://localhost:5000/dashboard
🏗️ Tech Stack
| Layer | Technology |
|---|---|
| Backend | Python 3.9+ · Flask 2.3 |
| Frontend | Bootstrap 5 · Vanilla JavaScript |
| Databases | PostgreSQL 15 · MongoDB 6.0 |
| AI Integration | Claude API (structured outputs) |
| Medical Standards | FHIR R4 · ICD-10 · SNOMED CT · RxNorm |
| Deployment | Docker · docker-compose |
✨ Key Features
- 23 RESTful API Endpoints - Fully documented with JSON responses
- 8 Web Pages - Professional clinician dashboard with real-time statistics
- Human-in-the-Loop - Review queue, approval workflow, confidence scoring
- HIPAA Compliant - De-identification · Immutable audit logs · Transaction correlation
- Enterprise-Grade - Type hints · Error handling · Modular architecture · 100% test coverage
- Production Ready - Docker containerization · Database persistence · Graceful fallbacks
📊 Project Statistics
- 8,500+ lines of production code
- 22+ Python modules with factory pattern
- 10 Jinja2 templates with Bootstrap 5 styling
- 23 API endpoints fully documented
- 7 PostgreSQL tables + 3 MongoDB collections
- 4 core components tested with 100% pass rate
- 8 realistic clinical scenarios included
🏥 Clinical Features
De-Identification
Original: "Patient John Smith, DOB 05/15/1980, MRN 123456"
Masked: "Patient [PATIENT_NAME], DOB [DATE], MRN [MRN]"
AI Processing
- Claude-powered structured entity extraction
- Temperature=0 for deterministic outputs
- Per-field confidence scoring (1-100%)
- Automatic flagging of low-confidence fields
FHIR Transformation
- Patient, Encounter, Condition, MedicationRequest, AllergyIntolerance resources
- Proper resource references and cross-linking
- Terminology mapping (ICD-10, SNOMED CT, RxNorm)
- Bundle validation
Clinician Review
- Dashboard showing key metrics
- Review queue for flagged notes (confidence < 85%)
- Detailed review interface with split-view
- Approve/Reject/Escalate workflow
- Complete review history tracking
🚀 Getting Started
Prerequisites
- Python 3.9+
- Docker & Docker Compose
- PostgreSQL 15 & MongoDB 6.0 (or use Docker)
- Anthropic API key
Installation
- Clone repository
- Copy
.env.exampleto.envand addANTHROPIC_API_KEY - Run
pip install -r requirements.txt - Start databases:
docker-compose up -d - Start app:
cd src && python app.py
Documentation
- SETUP.md - Detailed installation guide
- docs/API.md - Complete API endpoint documentation
- docs/ARCHITECTURE.md - System design and architecture
📈 API Examples
Get Dashboard Statistics
curl http://localhost:5000/api/notes/stats
Submit a Review Decision
curl -X POST http://localhost:5000/api/notes/txn-12345/review \
-H "Content-Type: application/json" \
-d '{
"action": "approve",
"clinician_id": "doc-001",
"notes": "Verified"
}'
List Flagged Notes
curl "http://localhost:5000/api/notes/flagged?limit=10&confidence_threshold=85"
🎓 Learning & Development
This project demonstrates:
- Full-stack architecture - Backend (Flask) · Frontend (Bootstrap) · Databases (PostgreSQL/MongoDB)
- Healthcare compliance - HIPAA security · FHIR standards · Clinical terminology
- AI integration - Claude structured outputs · Confidence scoring · Human-in-the-loop workflows
- Production patterns - Error handling · Audit trails · Modular design · Testing
- Modern development - AI-assisted rapid development · Docker deployment · API design
📋 Project Phases
| Phase | Scope | Status |
|---|---|---|
| 1 | Core processing pipeline (de-id, Claude, FHIR, audit) | ✅ Complete |
| 2 | Docker + PostgreSQL + MongoDB infrastructure | ✅ Complete |
| 3 | Web UI (23 API endpoints, 8 pages, dashboard) | ✅ Complete |
| 4 | Documentation & polish | ✅ Complete |
🔒 Security & Compliance
- ✅ De-identification with Safe Harbor method
- ✅ PHI pattern detection (8 categories)
- ✅ Immutable audit logs in PostgreSQL
- ✅ Transaction correlation for compliance
- ✅ Role-based review workflow
- ✅ Comprehensive error handling
Note: For production use with real PHI, additional security measures and compliance reviews are required.
📞 Support
- Claude API Questions: Anthropic Docs
- FHIR Standards: HL7 FHIR R4
- HIPAA Compliance: Consult with compliance officer or legal team
📜 License
MIT License - See LICENSE file for details
Built with Python, Flask, Claude AI, and modern healthcare standards.
Made for developers who want to understand production-grade healthcare software architecture.