Résumé

My work in backend development, infrastructure, team projects and open source.

Overview

I worked on education services and their deployment during full-time employment. Since then, I have completed two training projects, released two mobile apps and developed open-source tools under Theorvane.

Where Period Projects Role
Smart Dong School (employment) 2023.09 – 2025.11 Education services · LMS · infrastructure automation Backend owner (full-stack) · deployment & ops
SeSAC (training) 2026.06 – 2026.07 YouthPick — youth policy recommendation platform (team of 5) Team lead, full-stack
Side 2026.02 – present Maiary · Curora — two mobile apps Solo — from concept to operations
KOSA (training) 2026.04 – 2026.05 Final project — Proxmox + AWS hybrid infrastructure (team of 5) Infrastructure lead · architecture

Skills

Area  
Language Java 21, TypeScript / JavaScript, PHP, Dart
Backend Spring Boot 3.5, NestJS, Koa, Express, Apollo Server 4, TypeGraphQL, GraphQL, REST
Data MySQL, PostgreSQL, Redis, TypeORM, Prisma, JPA / Hibernate, DataLoader, Flyway, MinIO (S3)
Auth OAuth 2.0 implemented from scratch (Google · Naver · Kakao · Apple), JWT, Spring Security, role & permission modelling
Realtime Socket.IO 4, STOMP over WebSocket
Test JUnit 5 + H2, Mockito, @WebMvcTest / @DataJpaTest, Jest, Vitest
Cloud AWS (EC2, S3, VPC, NLB/ALB, Route 53, ACM, CloudFront, CodeDeploy, EKS/Fargate, Site-to-Site VPN, Client VPN)
IaC Terraform, Ansible, Packer
Container Kubernetes (kubeadm HA), Docker, Docker Compose, Helm, MetalLB, Ceph CSI
Virtualisation Proxmox VE (HA cluster), Ceph RBD, MinIO
Network / Security pfSense, VLAN segmentation, HAProxy + Keepalived VIP, WireGuard / IPsec VPN, nginx, IAM role design
CI/CD GitHub Actions, ArgoCD (GitOps), AWS CodeDeploy, PM2, blue/green zero-downtime deployment
Observability Prometheus, Grafana, Loki, Thanos, New Relic, Log4j2 → database sink
Frontend React 19, Next.js, Flutter, TanStack Query, Zustand, Riverpod
Practice Domain-driven package design, transaction boundary design, N+1 elimination, zero-downtime deployment, IaC, code review process

Open source — Theorvane

I develop MCP and LangChain tools, a GitLab client and a video editor under Theorvane, with public source code and usage examples.

  • TypeMCP — a TypeScript library for declaring MCP servers with decorators, with validation, MCP SDK integration, stdio and Streamable HTTP support.
  • TypeChain — a library for defining LangChain tools and agents with decorators, including a bridge to TypeMCP tools.
  • OpenScene — an Electron and React video editor whose AI agent can edit clips and start local FFmpeg exports.
  • LabFox — A Flutter GitLab client for issues, merge requests, code review, pipelines and job logs, with self-hosted and multi-account support. In development, ahead of store release.
  • Proxmox MCP — A TypeMCP server for the Proxmox VE API over stdio. Supports VM and LXC inspection and management, with explicit target confirmation for deletion and force-stop operations.
  • examples — a runnable example connecting a public REST API to an MCP server and a TypeChain agent.

1. Smart Dong School (employment)

2023.09 – 2025.11 (2 years 3 months) · Backend developer (full-stack) · deployment & operations · industrial technical personnel service — enrolled 2023.12, discharged 2025.11

I developed backend features and handled deployment and operations for a vocational training LMS and other education services. My work covered new features, maintenance and performance improvements.

Backend features and admin screens

TypeScript · PHP · Koa · GraphQL · TypeORM · MySQL · React

  • Built instructor settlement, notification configuration and external data integration features, along with their admin screens.
  • Worked on payment integration, exception handling and sensitive-data protection.
  • Developed real-time learning progress tracking and concurrent-session management.
  • Maintained campaign management, messaging and video-player features across education and marketing services.

Performance and legacy migration

New Relic APM · EXPLAIN · NestJS · DataLoader

  • Used APM and execution plans to investigate slow APIs. Improved queries with composite indexes, batched lookups and DataLoader.
  • Migrated Express code to NestJS while checking compatibility with existing APIs.
  • Moved repeated PHP code into shared modules and migrated screens to React in stages.
  • Upgraded React and Node in the admin console and revised its bundler and state management.

Infrastructure and deployment

AWS · Terraform · GitHub Actions

  • Wrote AWS infrastructure configuration in Terraform, choosing the scope after reviewing the impact on live services.
  • Built CI/CD and zero-downtime deployment procedures with automated health checks and recovery.
  • Set up environments for testing several features in parallel.
  • Added monitoring and static analysis tools and documented recurring procedures and handover notes.

2. SeSAC (training) — YouthPick

2026.06 – 2026.07 · SeSAC first team project · team of 5 · team lead, full-stack Spring Boot 3.5 · Java 21 · MySQL · Redis · MinIO · React 19 · Docker Compose · GitHub Actions

I developed authentication, admin, chat, file and profile features on the backend, plus the admin screens and frontend session handling. As team lead, I set the branch and review rules and configured deployment.

Backend and frontend

  • Social login — implemented OAuth 2.0 for three providers without a library. I handled callbacks and token issuance for an SPA on a separate domain, mapped provider responses to a common format and returned a shared error code for provider failures.
  • Token management — after introducing refresh-token rotation, I found that overlapping requests from multiple tabs or React StrictMode could cause 401 errors. I removed rotation and used hashed token storage, HttpOnly cookies and revocation on account suspension.
  • Error codes — defined a shared interface with codes implemented per domain. The frontend uses those codes to select messages for users.
  • Recommendation tests — a mismatch between profile enums and public-API mappings could produce a zero score without an exception. I added a test checking that the definitions agree.
  • Admin features — built the back office API and screens. The admin module depends on the domain modules, and authorisation rules are managed in one security configuration.
  • Real-time chat — aligned STOMP heartbeat settings across the client, broker and proxy to fix dropped idle connections. I put authorisation in a STOMP interceptor because the browser WebSocket API could not send the required handshake headers.

Team workflow and deployment

  • Configured Docker Compose to start the stack with one command. A proxy routes API requests to the backend, keeping browser requests on the same origin.
  • Set up GitHub Actions CI in both repositories. The backend runs formatting checks, a build and in-memory database tests. I configured cancellation of redundant runs, minimum token permissions and checkout without retained credentials.
  • Wrote an appender to store application logs in the database. The existing JDBC appender wrote empty strings into integer columns when values were absent, causing inserts to fail. I changed those bindings to NULL and made writes asynchronous, with separate handling for logging exceptions.
  • Added trace IDs, paths, IPs and user IDs to request logs and made warnings and errors searchable from the admin screen.

3. Side projects — Maiary · Curora

2026.02 – present · solo · shipped on the App Store · 32 cumulative release builds Flutter · Flame · NestJS · Prisma · PostgreSQL · Socket.IO · OpenAI

I designed, developed and released two mobile apps. I still operate both myself, so maintenance time and cost are part of each feature decision.

  • Subscriptions — verify receipts with Apple and Google servers and receive renewal, expiry and grace-period events through webhooks. I retain the original payloads for reprocessing.
  • LLM responses — constrain the response format with schemas and return defaults in the user’s language when a call fails. Cache expiry limits call volume, and cached errors prevent repeated failed calls.
  • Curora’s on-device storage — I moved RSS parsing and storage into the app to avoid maintaining a scheduled server collector. The server handles authentication, AI and payments. This allows one API server and offline reading, at the cost of cross-device sync.
  • Wrote RSS 2.0, Atom and RDF parsers, including BOM handling, RFC-822 dates and pagination.
  • Built iOS home-screen widgets with WidgetKit and Swift, sharing app data through an App Group. Widgets are a subscription feature.
  • Added a Redis adapter to Socket.IO multiplayer from the start so it could run across multiple server instances.
  • Support Korean, English, Japanese and Hindi. The admin console is generated from the schema.

4. KOSA (training) — final project

2026.04 – 2026.05 · KOSA Cloud Infrastructure Engineer Program · team GARDEN, 5 people · infrastructure lead Proxmox VE · Kubernetes · Terraform · Ansible · Packer · AWS

Built a four-server Proxmox cluster hosting 25 VMs and a Kubernetes HA cluster with 7 nodes and 106 pods, connected to AWS through VPN.

Network and high availability

  • Separated public access, DMZ and load balancing, internal services and management into four VLANs. Management and external access used a 1G network; Ceph storage used a separate 10G network.
  • Set up Keepalived VIPs for DNS, Vault, HAProxy, the Kubernetes API and monitoring. VMs with the same role were placed on different physical hosts.
  • Pointed the HAProxy VIP at the Kubernetes API VIP so changes to master nodes would not require load balancer edits.

Design decisions

  • Recovery services — placed CI/CD, Vault and MinIO on VMs outside Kubernetes. Keeping Terraform state inside the cluster would make recovery harder if the cluster failed.
  • Management access — placed the Proxmox management interface under the physical router so it would remain accessible if pfSense failed. I left pfSense redundancy out at this project scale and documented the reason and limitation.
  • Storage — used local volumes for Kubernetes masters to reduce etcd fsync latency, and Ceph RBD for the other VMs. Replication traffic used a dedicated 10G network with MTU 9000.
  • Air-gap transition — documented the sequence for completing Nexus mirroring before removing internet access through firewall rules. I did not validate the full transition and recorded that limitation.

Kubernetes platform

  • Deployed MetalLB, Ceph CSI, Harbor, ArgoCD, Gitea and Percona XtraDB Cluster with Helm, including address pools and StorageClass configuration.
  • Set PXC pod anti-affinity to requiredDuringScheduling to place its three replicas on separate nodes.
  • Handled Operator CRD ordering and MetalLB immutable-selector conflicts in deployment scripts.

AWS integration

  • Wrote nine Terraform modules for VPC, EC2, NLB, Route 53, key pairs, EKS, Site-to-Site VPN, WireGuard and CloudFront.
  • Connected pfSense to AWS through IPsec Site-to-Site VPN, matching encryption settings on both tunnels and configuring a parallel WireGuard path.
  • Used HAProxy to expose on-premise Kubernetes and AWS EKS through one endpoint, routing to EKS when on-premise health checks failed. Mirrored Harbor images to ECR and used ArgoCD to synchronise both clusters.
  • Designed cloud bursting that checks CPU usage in Prometheus and changes HAProxy ACLs to route traffic to AWS above a threshold.
  • Estimated Fargate costs from per-pod rates: about $180 per month for an active-active 70/30 split, compared with $120 in steady state and $150–200 during bursting.

Team results

Other members handled observability and security. The team retained metrics for over 30 days with Prometheus, Thanos and S3, collected node logs in Loki and deployed Wazuh SIEM to ten servers by script. Time from incident detection to a Gitea ticket fell from over 30 minutes to under 5.

Documentation

I wrote the design documents with the reasons for each choice and the alternatives considered. These covered Percona, Ceph RBD, Proxmox HA and DMZ separation, along with build guides for the other team members.


5. Home server and infrastructure operations

Proxmox VE · TrueNAS · ZFS · Kubernetes · GitLab · Wazuh

I designed and operated a Proxmox VE home server, separating services into VMs and LXC containers and managing their resources and networks.

  • Virtualisation — allocated CPU, memory and storage per VM or container and monitored resource usage and service health.
  • Storage — passed physical disks through to a TrueNAS VM for ZFS storage. Managed SMB and NFS shares, datasets and access permissions.
  • Kubernetes — configured a cluster with one master and two workers to deploy and operate applications.
  • Source control and CI/CD — hosted GitLab and build runners for source management and automated builds and deployments.
  • Self-hosted services — deployed and operated Nextcloud, Immich, Jellyfin, BookStack, a mail server and OpenCompany.
  • Network and security — kept internal services on private networks and used a reverse proxy, TLS and access controls for public services. Set up Wazuh for security-event monitoring and log analysis.
  • Operations — checked service health during incidents and validated releases before deployment, with procedures for returning to the previous version when needed.

Education

Period  
2017.03 – 2026.08 Sejong University — Computer Engineering, graduated

Training

Period Program
2026.06 – present SeSAC — team lead, first project (YouthPick)
2026.03 – 2026.05 KOSA Cloud Infrastructure Engineer Program — completed · infrastructure lead, final project

Certifications

Certification Issuer Date
AWS Certified Solutions Architect – Associate Amazon Web Services 2025.12.30
HashiCorp Certified: Terraform Associate (004) HashiCorp 2026.03.21
Engineer Information Processing HRD Korea 2025.06.13
SQL Developer (SQLD) Korea Data Agency 2025.04.04
TOEIC Speaking IH (150) YBM 2026.06.14

Awards

Award Date
3rd place, university SW coding competition 2020.12.11

Military service

   
Status Completed — industrial technical personnel, discharged on schedule
Placement Smart Dong School (designated company) — same as the employment above