Hello, dev.to — I'm Victor: 10+ years full-stack, two CTO runs, now solo and writing in the open

Hi. I'm Victor, a full-stack engineer based in South Korea, and this is my first post here.

I've been writing for a while — just in Korean, on my own blog. I'm starting fresh in English on dev.to because most of what I'm working on now (self-hosted LLMs, full-stack infra, building products as a one-person studio) lives in conversations that happen in English, and I'd rather be in those conversations than next to them.

So, a short introduction.

The path so far

I've spent 10+ years in this work. I started on the frontend, drifted into mobile, and eventually owned the whole stack — frontend, backend, infra, and the decisions that connect them.

Along the way I did two CTO runs:

• I led a small team to productize an education SaaS — architecting OCR, search, real-time messaging, and lecture streaming, and signing real customers. The thing actually became a business, which taught me more about tradeoffs than any side project ever did.
• I later owned the full technical architecture of another education platform — FE, BE, and DevOps, including the cloud setup and CI/CD pipeline.

More recently I spent a year as a remote full-stack contractor for a US SaaS (Wyzly), shipping features and running production on Next.js and Supabase across a 14-hour time difference. Remote-across-timezones is its own skill, and that year sharpened it.

I've also built enterprise AI work — an LLM agent system for a large organization — and earlier in my career, a stretch of interactive and kiosk experiences for major Korean enterprises and global brands at a creative-tech agency. Different worlds (white-glove enterprise vs. scrappy startup), and I'm glad I saw both.

What I'm doing now

These days I run a solo studio and spend most of my energy on two things.

Building my own products.

• Geckoly — an AI-assisted care platform for reptile keepers. (Yes, reptiles. It's a real, underserved niche with surprisingly hard problems — image-based species/price prediction, care guidance, the works.)
• BIoTan — peer-relative anomaly detection for IoT sensor fleets: instead of setting a threshold per device, it compares each device against its peers and flags the real outliers. No per-device tuning, common drift cancels out.

Running AI infrastructure on my own hardware.
A lot of my recent work is figuring out how far you can get on a single RTX 3090 — local model serving, an agent runtime that uses tools and cron and touches real files, KV-cache and quantization tuning, RAG pipelines. It's equal parts engineering and cost discipline, and it's the most fun I've had in years.

Why I'm here

I want to write in the open about the stuff I'm actually doing:

• Self-hosted LLM infra — what works, what melts the GPU, what the docs don't tell you.
• Full-stack notes — React Server Components, payments, the boring-but-real parts.
• Building solo — the engineering and the business side, honestly. Including the war stories (I have a good one about ransomware).

I'm also open to remote contracts and consulting with global teams — full-stack and AI-engineering work, or CTO-level architecture. But I'm not here to pitch; I'm here to be useful and to learn in public. If something I write helps you, that's the win.

If any of this overlaps with what you're working on — local LLMs, indie products, remote contracting — I'd genuinely like to hear how you're approaching it. Leave a comment, and I'll see you in the next post.

— Victor