I build systemsthat know whatto do next.
Most developers build features. I build systems that know what to do next. 17 years of production instincts, now multiplied by AI — I ship in days what agencies quote in months.
Systems that deliver.
Every project below runs in production. Real infrastructure, real users, real uptime — not portfolio demos.
Seoul.fm
liveFlagship17 years. 6 continents. $44/month infrastructure. The platform that never sleeps.
MessageBox
Three inboxes. One interface. Claude drafts. OpenClaw decides what to send autonomously.
AI CX System
A luxury retailer's entire CX operation — email triage, RAG chat, ops dashboard. Built in a weekend. A dev agency would quote $15k–$50k.
ListingLaunch
Property address in. HTML email + PDF flyer + social captions + Reels video out. Voice-trained on the agent's own content.
InboxAI
Every email read, classified hot/warm/cold, and drafted — before the agent even opens Gmail.
TV Karaoke System
The only K-pop platform with TV lyrics sync via Chromecast and AirPlay. 4 design iterations in 4 hours. Zero competitors.
Dual API Architecture
78 production endpoints across two architectures — built start to finish in one weekend.
SongIngest
6,400-song broadcast library, kept clean automatically. AI romanization, -16 LUFS normalization, Spotify metadata.
Integrated Home Infrastructure
18 cameras, 6 audio zones, VoIP, weather station, dual-WAN failover — all wired and installed from scratch.
Speed without shortcuts.
AI-native workflow that ships production features in hours, with the reliability you'd expect from months of development.
Think before you build
I spend more time in a doc than in an editor before touching code. Every constraint named, every edge case mapped. Rewrites are expensive. Clarity is free.
Multiply, don't shortcut
AI is a force multiplier — when you pair it with schema enforcement, validation layers, and the right human gates. I move 3–5× faster because the AI handles volume while I handle judgment.
Ship for the long run
Production means it runs at 3am without me. Seoul.fm has been self-sustaining for 17 years. I build every system to that standard — monitoring, error handling, and the reliability to be forgotten about.
Full-stack, AI-native.
The full picture — from a Claude prompt to a Proxmox VM, from a pgvector query to a Cast SDK receiver. Seventeen years across the stack means nothing is a black box.
AI & Automation
- ›Claude Code — primary dev environment
- ›RAG systems, pgvector, vector DBs
- ›n8n workflow orchestration
- ›OpenClaw agentic pipelines
- ›Prompt engineering & Claude Skills
- ›Google Gemini, OpenAI API
Backend Development
- ›Python (Flask, FastAPI, async, Redis/RQ)
- ›Node.js / Express
- ›PHP (modern, strict typing)
- ›PostgreSQL / MySQL / Supabase / Redis
- ›REST API design (78-endpoint systems)
Real-Time Systems
- ›Server-Sent Events (SSE)
- ›Redis pub/sub messaging
- ›Event-driven architecture
- ›<200ms push update latency
- ›HLS/CMAF adaptive streaming
Frontend & Mobile
- ›React / Next.js / Modern JavaScript
- ›Flutter — Android app on Play Store
- ›Progressive Web Apps (PWA)
- ›Tailwind CSS, shadcn/ui
- ›Remotion — programmatic video
Infrastructure
- ›Proxmox — multi-VM production hosting
- ›Docker / containerization
- ›Linux system administration (8+ years)
- ›FFmpeg / broadcast audio processing
- ›Cloudflare Workers, BunnyCDN
Approach
- ›AI-native development (3–5× faster)
- ›Autonomous systems design
- ›Production reliability focus
- ›Pieced-together systems thinking
- ›Ship it, learn, iterate
You have a problem. I have the tools.
Whether it's a workflow eating 40 hours a week, a system that needs to think for itself, or a product that needs to move fast — I've built for radio stations, real estate firms, luxury retailers, and my own 17-year platform. The pattern is always the same: the right AI stack + production-grade judgment = something that actually ships.