I turn repetitivecommerce operationsinto products
I'm Junseong Yun, a product engineer. I find the work that keeps repeating in day-to-day operations, define it as a problem, build a product for it and take responsibility for running it.
Right now I'm building the internal operations platform at LORENZ, a Korean pet brand, which brings orders, shipping, customer service and settlement into one place. On my own, I run CozySell, a subscription service I built that takes orders from 6 Korean marketplaces and carries each one through to purchase and payment at the supplier, a fashion platform.
What to expect from me
Each point is backed by work I've actually done.
I start from my colleagues' real questions
At LORENZ I gathered the requests our AI agent couldn't handle over 3 weeks. There were 30, and 13 of them were writes, so write actions that run after approval came next. I also compared 227 customer inquiry alerts with the answers staff actually gave and revised 9 internal wiki documents.
LORENZ caseI take responsibility for the whole service
I run every part of CozySell myself, from planning and development to subscription billing, deployment and customer servers. Since February 2026 that has meant 3,425 commits, and 241 releases in 60 days between June and early August 2026.
CozySell caseI prevent incidents with structure
CozySell's auto-purchase once paid again for an order an operator had already bought by hand. After that, uncertain cases are blocked instead of paid, and I added duplicate-guard and integration tests.
See the hard problemsI check improvements with numbers
I cut CozySell's production deploy from 38 minutes to 6 and customer image publishing from 10 minutes to 4. The Windows agent is now rebuilt only when its source changes, which cut rebuilds by about 80%.
See the resultsI build for colleagues who are not developers
At LORENZ, everyone on the 7-person team except the CEO was new to AI tools. I wrote plain-language install and user guides, and put announcements, feature requests and a bug board inside the admin screen.
How we workedI put AI into how the work gets done
Inside the company I added a Slack agent where every write needs a person's approval. For development I run several Claude Code and Codex agents in parallel, and I make the design, final review and release calls myself.
See key decisions
Experience
Dec 2025 to present
LORENZProduct Owner
I own the internal operations platform of a pet brand, which handles orders, shipping, customer service, settlement and marketing in one place. I decide what to build and do the design, development, deployment and staff training myself. I also built the brand's homepage for overseas buyers.
Django, Next.js, MCP server, Docker
Feb 2026 to present
CozySellProduct Engineer (solo)
I build and run a subscription back-office product for marketplace resellers on my own. Each customer gets a dedicated server or installs it on their own, and a license server I wrote handles subscription billing, deployment and updates.
Django, Next.js, FastAPI, AWS
2023 to present
EVERYTICKCTO
Under the EVERYTICK name I build and run Bdsillok, and I have taken on client work from online store skins and features to a CRM for an admissions consulting firm and a cafe operations system. I also built internal logistics automation and a B2B site.
Django, Next.js, PostGIS, Cafe24, Notion API
Case studies
For each project: what the problem was, what I weighed, how I solved it and what changed. Client names are replaced with their industry.
CozySell
Problem
Order collection was automated, but buying and paying at the supplier was still done by hand, one order at a time.
- Production deploy
- From 38 to 6 minutes
- Releases
- 241 in 60 days
LORENZ Operations Platform
Problem
Staff copied and entered orders one by one, and checked deliveries by hand on courier websites.
- Order hub
- 13 sales channels on one screen
- Ops agent
- 30-plus MCP tools, writes only after approval
Bdsillok
Problem
It was an old PHP site where the server redrew the whole result HTML on every calculation.
- Calculators
- 66
- Map
- 150,000 complexes in 150 clusters
Cafe Operations System
Problem
There was no way to manage inventory, and the owner was filing taxes without an accountant.
- Clock-in
- Only on the store's Wi-Fi or IP
- Store isolation
- A separate DB schema for each cafe
Admissions Consulting CRM
Problem
In peak season, many applicants had to be matched at once, so there was a high risk of booking a consultant twice for the same time.
- Consultation flow
- 9-stage state machine
- Server migration
- 12 scripts, verification written first
8,128 commits in the last year
This combines every repository I worked in. It includes commits made together with coding agents. I committed on 326 days of the year, and the longest streak was 188 days. The busiest day was Aug 25, 2026, with 100 commits. Most of them went into CozySell and the LORENZ Operations Platform.
- Nov 2025: 37 commits
- Dec 2025: 208 commits
- Jan 2026: 543 commits
- Feb 2026: 498 commits
- Mar 2026: 808 commits
- Apr 2026: 820 commits
- May 2026: 1046 commits
- Jun 2026: 800 commits
- Jul 2026: 851 commits
- Aug 2026: 970 commits
- Sep 2026: 756 commits
- Oct 2026: 743 commits
- Nov 2026: 48 commits
Other things I built
2026
Relay
A macOS app that hands Keychain secrets to AI agents for a fixed window only. Still in progress, with 122 test files.
SwiftUI, AppKit
TodoAll
Tasks, notes and an inbox in one app for macOS and iOS. 85 test files across 6 layers, from domain to UI.
SwiftUI, XcodeGen
DayOnRoutine
An iOS app that re-plans this week's workouts around your weight-goal pace. It reads weight records from Apple Health.
Swift 6, SwiftUI, HealthKit
Documentary production pipeline
A tool that produces long-form documentary videos in 13 stages, from research and script through fact-checking, narration, subtitles and rendering.
Bun, TypeScript, zod
Spatial inventory
An inventory tool for small businesses where stock sits on a floor plan you can walk through in 3D. Each business's data is separated with Postgres row-level security.
Hono, Drizzle, Next.js, Konva, three.js
EventMate
A prototype that collects prize-draw events from many sites, grades whether each one can be entered automatically and tells you. A monorepo of 3 apps and 4 packages, built in 78 commits over 3 days. The automatic entry step is only stubbed out so far.
NestJS, Next.js, tRPC, BullMQ
Marketplace product update extension
A Chrome extension that automatically re-runs product updates and transfers to 6 marketplaces inside a product management tool. Later merged into the CozySell extension.
Chrome extension (MV3)
POS order alerts
Reads the order count on the POS app screen and sends a Telegram alert when a new order arrives. It stops before sending any outside request until an official integration method is settled.
Python, Windows UI Automation
2025
Mobile wedding invitation
A mobile wedding invitation with an admin page, Naver Maps and KakaoTalk sharing.
Django REST, Next.js 15
Gmail to Notion collector
Reads Gmail every 3 hours and moves messages into a Notion database. Notion's 2,000-character limit is handled by splitting at sentence boundaries.
Python, Gmail API, GitHub Actions
Automatic content publishing
Claude writes copy for collected news posts, and only the ones picked in Telegram get posted to a YouTube community tab.
Django, Celery, Selenium, Claude API
YOLOv8 auto-labeling
The model labels first, people review only the low-confidence results, and the model retrains on them to grow the dataset.
Python, Ultralytics, ONNX
2024
EVERYTICK logistics automation
Every weekday it creates the day's order page in Notion with a sub-page per mall, and on the 1st and 16th of each month it creates the settlement request page. Weekends and holidays are skipped.
Python, Notion API, GitHub Actions
2023
Hair salon booking SaaS
Bookings, customers, staff schedules and text messages for hair salons in one place.
Django, Next.js
Tools I use
- Backend
- Python, Django, DRF, FastAPI, Celery, PostgreSQL, PostGIS, Redis
- Frontend
- TypeScript, Next.js, React, TanStack Query, Tailwind CSS
- Apps and extensions
- Swift, SwiftUI, Expo, Chrome extensions (MV3)
- Automation
- Playwright, Selenium, marketplace and online store APIs, Notion API, Telegram bots
- Infra
- Docker Compose, Nginx, AWS (EC2, ECR, Lightsail), Cloudflare, GitHub Actions
- AI
- Claude API, MCP servers, running coding agents (Claude Code, Codex)