AI Serving Platform
Redis Queue ingestion, GPU worker execution, Kubernetes scheduling, and result upload workflows.
Seoul, South Korea · AI/ML Infrastructure
Software Engineer building production AI infrastructure.
I design and operate Kubernetes-based AI/ML inference pipelines, GPU model serving systems, Redis queue workflows, CI/CD automation, and observability stacks for production medical AI workloads.
Event-Driven Autoscaling
KEDA-based scale-to-zero model workers to optimize GPU utilization and reduce idle cloud spend.
Production Operations
Prometheus, Grafana, cAdvisor, Node Exporter, Jenkins, ArgoCD, Helm, Docker, and Linux operations.
Experience
Current professional work
Software Engineer
JLK · Publicly traded medical AI company specializing in stroke analysis
- Designed and operated a production-grade AI inference pipeline serving 15 medical AI models on Kubernetes.
- Replaced always-on cloud containers with a scale-to-zero architecture, reducing monthly cloud cost by roughly 7M KRW.
- Configured KEDA to scale model workers from dedicated Redis queues only when requests arrive.
- Used Kubernetes PriorityClasses to prevent GPU starvation across default and scale-out workloads.
- Reduced cold-start pain from large CUDA images by triggering cluster-wide image pre-pulls through Jenkins and DaemonSet automation.
- Built production visibility with Prometheus, Grafana, Node Exporter, and cAdvisor.
Selected Projects
Recent infrastructure projects
Older school projects were removed so the portfolio focuses on current professional impact.
Production AI Platform
2024 — 2026
Kubernetes-based AI/ML inference pipeline
Built an infrastructure-agnostic model serving flow from request ingestion to Redis Queue, GPU worker execution, result upload, and monitoring for medical AI workloads.
Data Migration
Jan — Feb 2025
AWS S3 migration for medical imaging data
Migrated 250 million DICOM objects totaling 60TB from Ncloud Object Storage to AWS S3. Used Terraform, AWS DataSync, boto3 orchestration, and greedy sharding to balance parallel workers.
Deployment Optimization
2025
Cold-start reduction for CUDA image rollouts
Large CUDA-based images caused 5–10 minute cold starts. I automated cluster-wide image pre-pulling through a Jenkins-triggered DaemonSet to make model rollouts significantly faster.
Technical Skills
Stack I work with
Languages
Python, Java
Cloud & DevOps
Kubernetes, Docker, KEDA, Helm, ArgoCD, Jenkins, AWS DataSync, EC2, S3, CloudFront
Infrastructure
Linux, Redis, Nginx, GPU inference operations, CUDA-based containers
Backend
FastAPI, Celery, Gunicorn, Uvicorn
Observability
Prometheus, Grafana, Node Exporter, cAdvisor
Version Control
GitLab, GitHub
Education & Certification
Background
Virginia Tech
Bachelor of Science in Computer Science · Graduated Cum Laude
Certified Kubernetes Administrator
The Linux Foundation
Contact
Let’s build reliable AI infrastructure.
I’m open to backend, infrastructure, platform engineering, and AI/ML infrastructure roles where production reliability and system design matter.