JK JungLos Angeles · Est. 2006

I ship entireproducts.

Twenty years of production systems — MMO-scale databases, global commerce platforms, and AI-native SaaS. Lately: six live consumer products and an AI-operated ERP, each designed, built, and operated end to end.

View selected work→
GitHub·LinkedIn
Selected WorkIndex
20+
years building production systems
6
live products, solo-built end to end
10+
AI models running in production
0→1
founding engineer, three times

01 · Selected Work

Products that exist, in production, with users.

Everything below was designed, built, deployed, and operated end to end — architecture, backend, frontend, mobile, infra, and billing included.

LiveiOS & Android App · Solo dev · UX/UI by Aska Cheung

Garagey AI

A car spotting game played worldwide. AI identifies each car, and the server keeps score.

A car collecting game on the App Store and Google Play. On the phone itself, faces and text are pixelated and the car is cut out. Gemini identifies the make, model, specs, and value. The World Ranking scores every catch on the server, with three Cloud Run services and human review behind it. The iOS and Android apps are native Swift and Kotlin, released together.

SwiftUIKotlin · ComposeFirebaseCloud RunGeminiApple Vision · ML KitStoreKit 2 · RevenueCatAdMob
App StoreGoogle Playgarageyai.comCase study →
Garagey AI — garage collection and in-app camera
Grmpt design studio interface
LiveAI SaaS · Solo-built

Grmpt

AI fashion design studio — from text prompt to runway video.

A production SaaS where designers generate garment designs from text or sketches, composite virtual try-ons onto custom AI models, and animate the results into 1080p fashion film. Solo-built and operated: async FastAPI backend, Next.js frontend, Stripe metered billing, and multi-provider AI routing across Gemini, OpenAI, and Kling.

Next.js 16React 19FastAPIPostgreSQLStripeGeminiKling 3.0
grmpt.comCase study →
In ProductionCurrent Employer · Founding Lead

Project Caravel

An AI-operated commerce platform running a global fashion business — built as the first engineer.

The operating system of a global commerce company spanning the U.S., Korea, and China: a full ERP, sync workers federating 6+ marketplaces, SKU-level profitability intelligence, and AI agents that draft reorders and generate content. Built from zero as founding engineer. Name withheld — it runs a real company.

Next.jsTypeScriptPostgreSQLAWSGeminiOpenAILambda
Case study →
In DevelopmentMac & iPhone App · Solo dev

EvenHall

Run AI coding agents on your Mac. Direct and approve them from your phone.

A Mac app that runs Claude Code, Codex, and other coding agents in real terminals, paired with an iPhone app for giving instructions from anywhere. When an agent wants to edit files or run a command, the request goes to your phone and waits for you. The phone reaches the Mac over an end-to-end encrypted peer-to-peer link, so conversations and files stay on the Mac and no server keeps a copy. I built it solo, from the Rust core to both apps, backed by more than 6,000 automated tests.

RustFlutteriroh · P2PSQLitePTYClaude CodeCodex
In DevelopmentAI Wardrobe App · Solo-built

Style Match

Your closet, digitized — with real AI infrastructure behind it.

In active internal TestFlight testing — the product itself stays under wraps until launch, but the engineering underneath is real: an async FastAPI backend running a five-provider AI pipeline (Gemini, OpenAI, Replicate, on-device Apple Vision, and locally-loaded embedding models) turns a photo of a garment into a tagged, cut-out, studio-quality flat-lay; a pgvector-backed matching engine scores outfit compatibility and shopping decisions; and an SQS worker fleet with its own self-healing recovery layer keeps all of it off the request path. Solo-built, backend to client.

FlutterRiverpodFastAPIPostgreSQL · pgvectorGeminiOpenAIReplicate
Case study →
LiveiOS Game · Solo-built

Number Sliding Puzzle PvP

Up to four players race the same board in real time — coordinated by Redis, settled by the server.

A real-time multiplayer sliding-puzzle race on the App Store: 2–4 players, identical boards, combo attacks that disrupt opponents. A TypeScript backend owns matchmaking, Glicko ratings, and the credit economy, with Redis distributed locks and cross-node Socket.IO fanout keeping multi-instance state honest.

FlutterNode.jsTypeScriptSocket.IORedisFirestoreDocker
App StoreCase study →
Backlot rgb capture from UE5RGB
Backlot depth capture from UE5DEPTH
Backlot normals capture from UE5NORMALS
Backlot albedo capture from UE5ALBEDO
Open SourceOpen Source · ML Data Pipeline

Backlot

Turn any Unreal Engine 5 scene into pixel-aligned, ML-ready training data.

A synthetic-data pipeline running inside UE5's embedded Python: a synchronized four-camera rig captures RGB, float16 depth, world-space normals, and albedo with zero pose drift, plus per-frame intrinsics and 2D/3D bounding boxes — ingested into PostGIS and explored in a Next.js viewer with 3D trajectory playback.

PythonUE5FastAPIPostGISNext.jsReact Three FiberDocker
GitHubCase study →
Open SourceOpen Source · AI × Virtual Production

SetLab

From a text prompt to a production-ready 3D film set — deployed straight into Unreal Engine 5.

An AI-orchestrated virtual production pipeline: Claude lays out the scene as a validated spec, Rodin Gen-2 generates the meshes, FLUX and Gemini handle reference imagery, and an 871-line UE5 Python watcher places, textures, and lights the whole set — with natural-language edits applied in under a second.

PythonFastAPIClaudeRodin Gen-2FLUX 1.1Three.jsUE5 Python
GitHubCase study →
Number Strike Baseball — gameplay
Number Strike Baseball — battle log
Number Strike Baseball — your turn
LiveiOS Game · Solo-built

Number Strike Baseball

Real-time PvP number baseball with server-authoritative Elo — and provable anti-cheat.

A real-time, turn-based PvP deduction game on the App Store. Per-format Elo ratings, an 8-tier ladder, a credits economy, and a 1,000-bot opponent pool — with every rating, credit, and match result settled server-side, and an SHA-256 commit-reveal scheme that makes cheating provable.

FlutterFirestoreCloud FunctionsFCMFirebase AuthGitHub Actions
App StoreCase study →
Flashcards Alarm — alarm list
Flashcards Alarm — AI Smart Scan
Flashcards Alarm — alarm quiz
LiveiOS App · Solo dev · UX/UI by Aska Cheung

Flashcards Alarm

The alarm you can only silence by spelling the word correctly.

A vocabulary trainer fused with an alarm clock, live on the App Store in 8 languages. SM-2 spaced repetition schedules every word; Gemini powers Smart Scan (photograph a word list, import it all), AI definitions, and an adaptive placement quiz that builds a personalized starter pack.

FlutterRiverpod 3FirebaseGeminiCloud FunctionsRevenueCatSwiftUI WidgetKit
App StoreCase study →
LiveiOS Game · Solo-built

Snek Snacks

Classic snake, re-engineered — a frame-perfect arcade core wrapped in a 90-card collection gacha.

A retro NES-styled snake game built solo in Flutter + Flame, live on the App Store. Score is pizzas eaten; every pizza compounds the speed and the board auto-expands as you grow. Around the arcade core sits an 18-stage, 90-card collection gacha plus a full live-ops layer — AdMob, Firebase Analytics, and Game Center — with all art and chiptune audio procedurally generated in Python.

FlutterFlameDartFirebaseAdMobSoLoudGame Center
App StoreCase study →

02 · Why me

Three things I keep proving.

01

AI-native, in production

Not demos — billed products. Multi-provider model routing (Gemini, GPT, Kling) inside a live SaaS that users pay for monthly, AI agents with human-approval guardrails running a real company's operations, and Claude-orchestrated 3D pipelines in open source.

02

Scale was the first job

Before AI, the training ground was live gaming infrastructure: 12 primary MySQL databases, 100M+ rows per table, 500GB monthly growth, failover cut from 10 minutes to under 60 seconds.

03

Zero to one, repeatedly

Founding engineer at three companies, the last two back to back. ERP systems that replaced paper, marketplaces that launched in new countries, engineering teams hired and mentored — and the architecture decisions that let all of it keep shipping.

03 · Capabilities

The stack, as actually used.

Every item below appears somewhere in the projects above — nothing aspirational.

AI / Agentic

  • LLM integration (Claude, GPT, Gemini, Kling, Veo)
  • AI agent orchestration
  • MCP
  • RAG pipelines
  • Multi-provider model routing
  • Embeddings (pgvector, CLIP-family)

Languages

  • Python
  • TypeScript
  • JavaScript
  • SQL
  • Dart
  • C++
  • C
  • Java
  • PHP
  • ASP

Backend

  • FastAPI (async)
  • Next.js
  • Node.js
  • SQLAlchemy 2.0
  • OAuth2 / JWT
  • REST APIs
  • Socket.IO

Data

  • PostgreSQL
  • MySQL
  • PostGIS
  • pgvector
  • Redis
  • Firestore
  • Supabase
  • MongoDB

Cloud & Infra

  • AWS (EC2, RDS, S3, CloudFront, ALB, SQS, Lambda, MediaConvert, ECR, CodePipeline, CodeBuild, CodeDeploy, CloudWatch, STS, SSM)
  • Docker
  • Terraform
  • nginx
  • CI/CD

Mobile

  • Flutter
  • React Native
  • Riverpod
  • SwiftUI WidgetKit
  • App Store deployment
  • RevenueCat

Payments & Trust

  • Stripe (subscriptions, webhooks, proration)
  • Credit-ledger billing
  • Audit logging
  • Row-level security
  • AES-256-GCM field encryption

Quality & Ops

  • pytest / Jest / Mocktail
  • Chaos testing
  • Firestore rules testing
  • Sentry
  • CloudWatch
  • gitleaks

Recognition

  • The President's Volunteer Service Award
    Barack Obama, 44th U.S. President
  • Certificate of Recognition
    Kevin de León, California State Senate
  • Certificate of Appreciation
    Eric Garcetti, Mayor of Los Angeles

Languages

  • KoreanNative
  • EnglishFull professional
  • JapaneseFull professional

04 · Contact

Looking for someone who ships?

I'm based in Los Angeles and open to staff-level engineering and founding-engineer roles. The fastest way to evaluate me is to use the products above — then let's talk.

jaykaysm@gmail.comLinkedInGitHub