I currently work as an AI Service Planner at Keduall, currently available for work.
Focused on AI learning services and Korean NLP, working from Seoul, South Korea.
I specialize in planning AI-powered learning services for Korean education — turning vague ideas into problem data, feature specs, and working prototypes. I'm passionate about building software that makes a difference.
Beyond my work as an AI service planner at Keduall, I run SeolGA, a micro product studio where I build tools like TextViz, OutlineForge, Promptlog, and FunCommute — combining product planning with hands-on AI engineering.
Trained as a Korean linguistics scholar with a Hallyu software double major, I treat Korean as both language and research material — building NLP datasets, annotation standards, and education services around it.
I'm currently working on some exciting projects that I can't wait to share with you. But I'm always open to new opportunities and collaborations.
PRD, IA, UI/UX Documentation, MVP Scoping, Feature Specifications, User Flows, User Research, Interview Design, Waitlist Landing Planning.
Prompt Engineering, LLM Evaluation, RAG, Multi-Agent Systems, OCR Pipelines, Data Augmentation, Question Generation, Golden Set Construction, Python.
TOPIK & Korean Education Domain, Korean NLP, TypeScript, Next.js, Tauri, Figma, Cursor, Claude Code, Git/GitHub.
Always exploring new tools, models, and ways to turn scattered ideas into working products :)
From AI learning services to developer tools — every project here connects planning, data, and working prototypes.
An AI writing tutor for TOPIK learners, planned end to end at Keduall — MVP feature specs, IA/UI/UX documents, and user flows for the 51–54 writing question types, delivered with working prototypes to the development team.
A question generation and validation pipeline for AI-powered TOPIK practice — rule-based generation, golden-set review criteria, and 100,000+ expected user queries with 22,000+ feedback records.
Prototypes for taking Korean education content global — an ePub multilingual translation tool with LLM re-review flows, and an OCR pipeline tested for digitizing textbook PDFs, validated against Vietnamese translation quality benchmarks.
A unified web editor from my studio SeolGA that handles Markdown, Mermaid diagrams, and LaTeX math in one workspace — with a JSON Prompt Builder that assembles reusable prompt fragments into structured results.
A CLI-first design quality gate for AI-generated interfaces — catching generic layouts, weak contrast, and "AI slop" patterns before they ship, built from hundreds of hours of shipping with coding agents.
My capstone project: a Korean learning web game site with the Daejeon KKumdol-i mascot — my first web service in the Korean education domain.
I run SeolGA, a micro product studio turning complex thoughts into practical tools — from planning documents and diagrams to prompt libraries and reports.
Products include TextViz, OutlineForge, Promptlog, DeckForge, and FunCommute.
The beliefs that shape how I plan, build data, and ship products
The human is the bottleneck. The best technology multiplies what people can do — it doesn't just replace them. Every tool I build is aimed at widening what one person can carry.
Why I build tools
Define the problem precisely before reaching for a tool. I like finding the connections between data — what evidence is needed, and what criteria will judge the result — before anything gets built.
Before the tool
Turn vague requirements into screens, data, and workflows. Documents alone don't ship products — I deliver HTML wireframes and working prototypes that developers can actually see and run.
Ideas into artifacts
Judge results against a golden set, not against taste. From resume data to generated TOPIK questions, evaluation criteria are what turn opinions into decisions and iterations into progress.
Evidence over taste
Touch new technology first. Possibility and limits only become real when you build with it — I prototype with the newest models and frameworks, then check whether they survive real cost and market conditions.
New tech, verified
Vertical domain knowledge wins. In AI, the difference is made by deep domain expertise — for me, Korean education and NLP — combined with the speed to verify ideas end to end.
Korean education × AI
Got a question, proposal, project, or want to work together on something?
