I build AI-powered services and the development workflows behind them. From product planning to backend engineering, release, and operations.

Featured Projects

K-Bap

K-Bap

Product planning & full-stack development / 2 developers / 2026, ongoing

A mobile app that reads Korean menus and explains food risks based on a visitor’s allergies and dietary preferences. Available on Google Play and the App Store in 10 languages.

App captures from October 14, 2026. Select a screen to view it at full size.

10,000 downloads
Android and iOS combined
10
Supported languages, including Korean
37 / 92
Initial requirements / development tasks
React Native / ExpoKotlin / Spring Boot MySQL / RedisLLM OCR Claude Code / CodexPrometheus / Grafana

My role

I owned product planning and worked on the mobile app, backend features, and operations. My teammate built the initial product AI pipeline; I later took over its maintenance and improvements. The AI workflow below is the process I designed for my own work.

How I work with Claude Code and Codex

I define scope, priorities, acceptance criteria, and review boundaries before implementation. Role-based sessions share API contracts, designs, and task records; a command-center session coordinates dependencies and handoffs.

  1. Define the work Requirements, API contracts, design references, and acceptance criteria
  2. Build and review separately Role-based sessions, exact-commit reviews, and device checks
  3. Decide and follow through Human decisions, release verification, and scheduled follow-ups
Coordinating a shared Figma workspace

Built a local Figma connection and ownership protocol with Codex. Planning can proceed in parallel, while editing and export require one owner. A unique claim, heartbeat, and explicit release prevent an old session from releasing a new session’s lock.

In an isolated test, 1 of 12 concurrent requests acquired the lock. Applied to design work; this is a cooperative session protocol, not an OS-level lock.

Verifying a cleanup job beyond the happy path

I defined the preservation policy and review scope. Independent AI review found that protected photos could repeatedly consume a cleanup job’s limit, starving ordinary deletion candidates. The implementation session added a Redis scan cursor.

Isolated MySQL, Redis, and local S3 substitute: 7 protected photos + 1 ordinary photo, with deletions of 0, 0, then 1 across three runs. Production-wide behavior was not inferred from this test.

Scheduled work with a completion contract

Designed a daily Codex workflow for research, drafting, independent review, publication, and public-page verification. Date-based records prevent duplicates; repeated failures stop the workflow. A separate PR watcher compares review commits, open findings, and CI without merging.

As of Oct 22: 6 daily runs; 12 reviewed drafts and 11 published articles including the initial batch. Claude Code card-news routines separately produced 31 draft records by Oct 20, not 31 verified social posts.

AI features and release operations

  • Review translation: versioned caches using the source hash and prompt version; no DB connection held while waiting for the model. Bounded concurrency and a 12-second timeout constrain resource use.
  • Applied canary deployment and Prometheus/Grafana monitoring. Distinguished OTA updates from native rebuilds, and checked the actual store submission artifacts.
  • For additive schema changes, separated the migration from the feature rollout to account for API migration execution and batch schema validation.

DuckOn

Team lead, backend & DevOps / 5 people / Launched Aug 2025; operated for ~10 months

A video co-viewing and chat service for global K-pop fans. I led the team and worked on backend development and infrastructure. Reached 1,000 MAU across 10 countries; operated for about 10 months before shutting down in mid-2026.

Java / Spring Boot MySQL / Redis Docker / GitLab CI/CD NGINX Reverse Proxy AWS EC2 / S3 / CloudFront OpenAI GPT API
93%
Average TTFB / LCP improvement
1,000 MAU
Users across 10 countries
10 months
Hands-on service operations

Key contributions

  • Designed and operated AWS infrastructure (EC2, S3, CloudFront, Route 53) with separate prod/dev environments and environment-specific CI/CD pipelines
  • Migrated static assets to S3 + CloudFront, configured OAC for security, and added cache-invalidation scripts to the CI pipeline
  • Integrated OpenAI translation on user request to avoid unnecessary model calls
  • Built real-time monitoring with CloudWatch, Lambda, SNS, EventBridge — automated error alerts and daily infra reports (CPU, memory, disk, container status) via Mattermost
  • Detected disk usage approaching its threshold in the daily report and addressed it before deployment
DuckOn architecture evolution
Evolution of the service infrastructure
DuckOn infrastructure diagram
Full AWS infrastructure — separate prod/dev, CDN, monitoring
CloudFront performance metrics
TTFB / LCP before-after CloudFront migration
Monitoring dashboard
Real-time alerts & daily infra reports via Mattermost
AI translation chat
OpenAI GPT-powered real-time translation chat
MemeDuck platform
MemeDuck — meme upload, search, browsing

Experience

WIPS — Frontend Developer Intern

Participated in a new in-house CRM system build, implementing customer-management screens and data-retrieval features with Vue.js.

Worked closely with planning and infrastructure teams through weekly meetings to gather requirements and reflect real-world feedback into UI improvements — full exposure to the product development lifecycle from requirements to iteration.

Activities

Developer, YAPP

Built an Android WebView screen with React and Recoil.

Developer, CMC

Shipped a hybrid app built with React Native and Redux Toolkit.

Frontend Developer, WoozLabs (Monomail)

Built a React/Electron desktop app for Gemini-assisted email drafting — signature management plus AI-powered tone/summary editing.

Awards & Honors

Samsung SW AI Academy Special Project – 2nd Place

Developed a banking application for foreign residents in Korea.

Samsung SW AI Academy Joint Project – 1st Place

Built a real-time video co-viewing & chat service, reaching 1,000 MAU.

Inha University Capstone Design Fair – 2nd Place

Developed a mutual braille/text translation application.

Silicon Valley Koreans Hackathon – 2nd Place

Built "Know Your Neighbors," a map-based regional safety-info sharing service.

IGRUS IT Competition – 1st Place

Built a Unity-based 3D game introducing the CS department to incoming freshmen.

Inha University Ideathon Competition – 2nd Place

Proposed an AR/VR character-restoration idea using photos and videos.

Education & Certifications

Education

Inha University
B.S. Computer Science | 2019 - 2025
SSAFY — Java Track
Samsung SW AI Academy | Jan 2025 - Dec 2025
Taylor University — Exchange Student
Upland, Indiana, USA | Jan 2023 - May 2023

Certifications

AWS Solutions Architect – Associate
AWS | Sep 2025
SQLD – SQL Developer
Korea Data Agency | Sep 2024

Languages

OPIc — AL (Advanced Low)
ACTFL | Feb 2025
TOEIC — 945 / 990
ETS | Sep 2024

Technical Skills

AI development & collaboration

  • Claude Code / Codex
  • LLM API / OCR / translation
  • Role-based sessions and independent review
  • Scheduled workflows and tool integration
  • React Native / Expo

Backend & APIs

  • Java / Kotlin Backend services
  • Spring Boot — main API framework
  • JPA / Hibernate — ORM
  • REST API design
  • JWT / OAuth

Cloud & DevOps

  • AWS — EC2, S3, CloudFront
  • Docker — containers & deployment
  • GitLab CI/CD — automated pipelines
  • NGINX — reverse proxy, SSL, load balancing
  • Canary deployment / Prometheus / Grafana

Databases

  • MySQL — relational, structured data
  • Redis — caching, sessions, pub/sub
  • Query optimization & indexing

Get In Touch

Emailqja086351@naver.com
Phone(+82) 10-8399-7482