About

I'm an engineer who likes taking a vague problem and turning it into something that works, fast. I studied Electronic & Electrical Engineering at Sungkyunkwan University and spent time in both industry and research labs along the way: NPU architecture for Transformer models at Samsung S.LSI, sim2real reinforcement learning at KAIST, GAN-based SoC placement at POSTECH, and computer-vision pose estimation at SKKU.

Then I started building. naly was an AI companion speaker, and I went from zero hardware experience to a custom Linux OS, a multi-layer PCB, and a week spent debugging audio glitches on a factory line in Shenzhen, all in under ten months. It taught me what hardware actually demands, and how to kill a project on time.

mefriend.ai came next, an AI character platform that scaled to ~300K users. I moved the hot path from Python to Go while it was under load, in a week, and cut LLM inference costs by about 80% with RAG, caching, and model routing. That last part only worked because I'd built an eval harness first. Desk of Einstein is an infinite-canvas workspace with a Figma-class extension platform I designed on my own. And claude-devtools, a weekend tool that somehow grew to 3.5K+ stars and 90K+ downloads, was the one where the distribution turned out to be the real engineering.

The pattern is the same every time: take a messy problem, keep the whole system in my head, and ship. I've worked the full stack of a hard product, from silicon and custom Linux up through distributed AI infra, developer platforms, and the customers using them — most recently as an AI Engineer on IBM's Client Engineering team, building agentic-AI systems and RAG/LLM pipelines for enterprise deployments. What I want next is to go deeper on the hardest parts of systems like these, on a team or problem worth owning end to end.

You can find my work on GitHub, connect on LinkedIn, or grab my résumé.