CLOUD / INFRA ENGINEER

I design and run systems
so services keep running.

I built and ran hybrid AWS–on-prem infrastructure over VPN, Kafka-based real-time pipelines and GitOps-managed Kubernetes, all hands-on.

BARO · vehicle position flow
Site-to-Site VPNVehicles ×1500asyncio simulatorMosquittoMQTT · EC2controlshared subKafka4 partitionsdispatchGEO dispatchTimescaleDBon-prem storePrice feedsyfinance · KIS WSKafkastock.raw.*Detection% · Z · ETFNewsNewsAPI · RSSLLM analysisGroq · DLQDashboardWebSocket · Slack

The core flows of both projects play in turn

1,500

concurrent simulated vehicles
handled over MQTT → Kafka

BARO

2.9M

Kafka consumer lag
traced to its cause and cleared

BARO

26

Kubernetes workloads
run with ArgoCD GitOps

StockPulse

108

Korean and US tickers
tracked by the anomaly pipeline

StockPulse

Projects

Site-to-Site VPNVehicles ×1500asyncio simulatorMosquittoMQTT · EC2controlshared subKafka4 partitionsdispatchGEO dispatchTimescaleDBon-prem store

BARO — Autonomous Taxi Dispatch & Control Platform

Apr–Jun 2026 · Team of 5 · Backend / DevOps

A hybrid cloud that ingests positions from 1,500 vehicles over MQTT → Kafka to dispatch and relocate them, running AWS services alongside on-prem analytics and observability linked over VPN.

  • AWS ECS
  • Terraform
  • MQTT
  • Kafka
  • Valkey
  • OpenStack · K3s
  • Airflow
  • Prometheus
Case study →
Price feedsyfinance · KIS WSKafkastock.raw.*Detection% · Z · ETFNewsNewsAPI · RSSLLM analysisGroq · DLQDashboardWebSocket · Slack

StockPulse — Stock Anomaly Detection & AI Analysis Pipeline

Solo project · Design to operations

Streams 1-minute bars for 108 Korean and US tickers through a Kafka pipeline to catch sharp moves, has an LLM explain the cause, and delivers it to a dashboard and Slack. Runs on an on-prem Kubernetes cluster, deployed with ArgoCD GitOps.

  • Kafka
  • Kubernetes
  • ArgoCD
  • Sealed Secrets
  • TimescaleDB
  • Prometheus
  • FastAPI
  • Next.js
Case study →

BARO

When Kafka stalled, the control console stalled too

Symptom
The control console showed only one vehicle, and Mosquitto logs filled with “Broken pipe” errors.
Cause
The Kafka producer’s default wait (max.block.ms, 60 s) blocked the MQTT receive thread, and the broker’s queue overflowed.
Fix
Cut max.block.ms to 500 ms and retries to 0 so a Kafka outage can no longer spill into MQTT ingestion.

BARO

Consumer lag hit 2.9M — the culprit was a DB query on the hot path

Symptom
The dispatch service fell behind on positions, and Kafka consumer lag grew to 2.9M.
Cause
For every position message, it looked up the vehicle’s active dispatch in the database.
Fix
Cached the vehicle → dispatch mapping in memory so message handling no longer touches the database.

StockPulse

ArgoCD stuck in OutOfSync forever

Symptom
The StatefulSets stayed OutOfSync after every deploy.
Cause
Fields the cluster fills in, such as StatefulSet volumeClaimTemplates, never matched Git.
Fix
Excluded those fields with ignoreDifferences, along with CronJob status for the same reason. ServerSideApply=true had been turned on in advance so large resources would never hit the 262 KB limit of the last-applied annotation.

Tech stack

Cloud · AWS

  • ECS Fargate
  • EC2
  • RDS
  • ElastiCache
  • ALB
  • Route53
  • Site-to-Site VPN
  • Secrets Manager
  • CloudWatch

IaC · CI/CD

  • Terraform
  • GitHub Actions
  • self-hosted runner
  • CodeDeploy Blue/Green
  • ArgoCD
  • Sealed Secrets

Container · Orchestration

  • Kubernetes
  • K3s
  • OpenStack
  • Docker
  • Harbor
  • MetalLB
  • NFS

Messaging · Data

  • Kafka
  • MQTT · Mosquitto
  • TimescaleDB
  • PostgreSQL · PostGIS
  • Valkey · Redis
  • Airflow

Observability

  • Prometheus
  • Grafana
  • Alertmanager
  • CloudWatch exporter
  • blackbox exporter
  • kafka-exporter

Backend

  • Kotlin
  • Spring Boot
  • Python
  • FastAPI
  • TypeScript
  • Next.js

About

I have built and run infrastructure across AWS and on-prem, hands-on. In BARO, the team project for the cloud track of Hyundai AutoEver’s Mobility SW School (cohort 3), my role was backend and DevOps, working on a hybrid setup that links AWS (ECS, Terraform) and an on-prem OpenStack K3s cluster over a Site-to-Site VPN. As we scaled up the vehicle simulation, I traced problems such as Kafka consumer lag and Mosquitto queue limits back to their causes and fixed them.

StockPulse is a solo project I took from design to operations. I put a Kafka-based pipeline on an on-prem Kubernetes cluster and set up a GitOps flow with ArgoCD and Sealed Secrets, where a commit to Git becomes a deploy.

When something breaks, I confirm the cause in logs, metrics and code rather than guessing, and once it is fixed I capture the fix in configuration and procedures so it does not happen again. What didn’t work is written down in each project’s limitations and retrospective, too. This site itself runs on S3 and CloudFront built with Terraform and deploys through GitHub Actions (OIDC).