curma
AI / ML engineer and Data Scientist with a B.S. in Cognitive Science, Machine Learning & Neural Computation.
Experienced in building production-grade GenAI solutions, including RAG pipelines and recommendation systems deployed for major Korean financial and hospitality platforms.
Independently designs and ships endto-end data intensive projects spanning generative probabilistic models, civic tech platforms, and quantitative trading systems.
Fluent in English, Korean, and conversational French and Chinese. Track record of working across Korea and the US in fast-moving startup and enterprise environments.
OKX Monthly competition Top 50 (BTCÐ to the moon)
Tencent API exposure discoverer
Experienced in building production-grade GenAI solutions, including RAG pipelines and recommendation systems deployed for major Korean financial and hospitality platforms.
Independently designs and ships endto-end data intensive projects spanning generative probabilistic models, civic tech platforms, and quantitative trading systems.
Fluent in English, Korean, and conversational French and Chinese. Track record of working across Korea and the US in fast-moving startup and enterprise environments.
OKX Monthly competition Top 50 (BTCÐ to the moon)
Tencent API exposure discoverer
Experience: 1 year
Yearly salary: $100,000
Hourly rate: $50
Nationality: 🇰🇷 South Korea
Residency: 🇺🇸 United States
Experience
Developer
PoliScope 2026 - 2026
> Civic transparency platform tracking 286 National Assembly members across bill proposals, votes, and attendance. > Built with Next.js, TypeScript, Supabase (PostgreSQL), Upstash Redis, and D3.js + TopoJSON for interactive district maps. > 18,128 bills · 1,548 recorded votes · AI summaries on ~1,360 passed laws · data refreshed daily via automated pipeline. › AI curation that ranks bills by impact per user type (workers, students, self-employed, etc.)
Research
2206 DRIFT 2026 - 2026
> DRIFT (Dynamic Rating and Influence Form Tracker) > The first generative state-space model for tracking professional League of Legends player skill form over time. > Decomposes performance into 5 latent dimensions, parameter estimation via EM algorithm, win prediction via Kalman Filter, and career trajectory visualization via RTS smoother. › Designed end-to-end data pipeline covering LCK, LPL, LEC, LCS across 121,070+ games > Current iteration achieves 4-5x improvement in Expected Calibration Error vs. TrueSkill and PandaSkill. > Accidentally discovered Tencent API exposure to backend data platform (reported!)
Data Science
GenON 2023 - 2023
> LLM-based financial document summarization and Q&A system for Woori Bank using retrieval-augmented generation and prompt engineering, training data augmentation with open-source models to deliver accurate, user-driven answers. › LLM-powered recommendation assistant for GoodChoice, Korea's leading hotel reservation platform, improving search narrowing and surfacing personalized hotel recommendations. › Reduced hallucinations in model outputs through systematic research into temperature tradeoffs and promptengineering, building rigor for shipping reliable on-premise AI.
Skills
ai
big-data
crypto
machine-learning
python
english
french
korean