Healthcare Data Platform
🏥 Overview
Representative work from a healthcare consulting environment: a data platform connecting operational, appointment, and advertising data for Japanese clinics. I built reporting workflows with Python, Airflow, SQL, GCP, and BI tools, reducing executive-report preparation from five hours to 30 minutes.
The work started with individual client reporting needs and expanded into reusable data patterns across clinics. Automated reservation and billing pipelines supported 40+ clinics, while shared KPI dashboards helped one major dental group grow revenue from approximately JPY 600M to JPY 800M.
Tech Stack
This project is separated into 4 components frontend, backend, infrastructure, and ETL server.
- Frontend
- Nextjs
- Ant Design
- Backend
- Django
- Django-Rest-Framework
- Database / DWH
- PostgreSQL
- Redis
- Snowflake
- Infrastructure
- AWS
- ECS Fargate
- Elasticache
- RDS
- S3
- Application Load Balancer
- IaC
- Terraform
- AWS
- ETL
- Apache Spark
- Apache Airflow
- Docker
- Python
- CI/CD
- GitHub Actions
📊 Technology Selection Rationale
Data Lake Comparison
| Feature | Apache Iceberg | Delta Lake | Apache Hudi | Traditional Parquet |
|---|---|---|---|---|
| ACID Transactions | ✅ Full | ✅ Full | ✅ Full | ❌ None |
| Schema Evolution | ✅ Excellent | ✅ Good | ✅ Good | ⚠️ Limited |
| Time Travel | ✅ Built-in | ✅ Built-in | ✅ Built-in | ❌ None |
| Partition Evolution | ✅ Dynamic | ⚠️ Static | ⚠️ Static | ❌ None |
| Multi-Engine Support | ✅ Spark, Presto, Flink | ⚠️ Spark-focused | ⚠️ Spark-focused | ✅ Universal |
| Healthcare Use Case | ✅ Best fit | ✅ Good | ✅ Good | ❌ Limited |
ETL Framework Comparison
| Feature | Apache Spark | Apache Flink | Traditional ETL | Python Scripts |
|---|---|---|---|---|
| Scalability | ✅ Excellent | ✅ Excellent | ⚠️ Limited | ❌ Poor |
| Batch Processing | ✅ Excellent | ✅ Good | ✅ Good | ✅ Good |
| Stream Processing | ✅ Good | ✅ Excellent | ❌ None | ❌ None |
| Healthcare Data Volume | ✅ Handles TB+ | ✅ Handles TB+ | ⚠️ GB scale | ❌ MB scale |
| Learning Curve | ⚠️ Moderate | ⚠️ Steep | ✅ Easy | ✅ Easy |
| Community Support | ✅ Large | ✅ Growing | ⚠️ Vendor-specific | ✅ Large |
Backend Framework Comparison
| Feature | Django + DRF | FastAPI | Spring Boot | Node.js |
|---|---|---|---|---|
| Development Speed | ✅ Fast | ✅ Fast | ⚠️ Moderate | ✅ Fast |
| Type Safety | ⚠️ Optional | ✅ Built-in | ✅ Built-in | ⚠️ Optional |
| ORM | ✅ Excellent | ⚠️ External | ✅ Good | ⚠️ External |
| Admin Interface | ✅ Built-in | ❌ None | ❌ None | ❌ None |
| Healthcare Compliance | ✅ Mature libs | ⚠️ Growing | ✅ Mature | ⚠️ Variable |
| Team Expertise | ✅ High | ⚠️ Learning | ⚠️ Low | ⚠️ Medium |
🚀 Next Steps
- Phase 1: Data Collection
- Integrate with receipt computer systems (レセコン) in pilot clinics
- Establish secure data pipelines
- Implement data quality validation framework
- Phase 2: Analytics Enhancement
- Develop predictive models for patient volume forecasting
- Build revenue optimization algorithms
- Create benchmarking system with anonymized peer data
- Phase 3: Scale and Expand
- Onboard 50+ clinics in the Tokyo metropolitan area
- Add support for specialized clinics (dental, dermatology)
- Launch mobile app for clinic administrators
- Phase 4: AI Integration
- Implement natural language processing for unstructured clinical notes
- Deploy recommendation engine for operational improvements
- Introduce automated anomaly detection for billing errors