Deep dives into how we solve complex technical problems. From healthcare systems to AI platforms.
AI-powered platform for speech improvement with real-time analysis and personalized coaching. Built with Next.js and LLM APIs for speech pattern recognition. The system analyzes pace, clarity, and emotional tone, providing instant feedback and progress tracking through an intuitive dashboard.
Enterprise medical imaging platform processing 100GB+ DICOM data daily for US telemedicine providers. Microservices architecture with FastAPI, Redis Streams for audit trails, and HAProxy load balancing. Handles routing from multiple clinics with HIPAA compliance, zero data loss, and real-time monitoring via React dashboard. Scales from single-clinic to distributed multi-site deployments.
Self-hosted GPU-powered Jupyter platform for AI education, deployed in Korean national research project. NestJS GraphQL API manages users and sessions, FastAPI controls Docker containers and GPU allocation. Students get isolated Tesla V100 GPU access via NVIDIA MIG. Supports 40+ concurrent users with real-time monitoring (Prometheus, Grafana) and replaces commercial cloud services for on-premise AI training.
Secure cloud-native DICOM backend for DXA scan storage and retrieval. Built with Orthanc DICOM server, AWS S3 for storage, PostgreSQL for metadata, and comprehensive CloudWatch monitoring. Features VPN-only access, HTTPS encryption, automated daily backups, and production-grade observability.
Enterprise PACS platform modernization for multi-site medical imaging center. Built custom web interface with Next.js and Ant Design, NestJS backend, Orthanc DICOM server, PostgreSQL metadata indexing, and Redis caching. Features advanced filtering, batch operations (anonymize, ZIP, send), role-based access control integrated with RIS, OHIF viewer integration, and WebSocket real-time progress updates.