Designing and shipping production-grade AI / data systems — model, service, pipeline, infra.
End-to-end builder. I take ML/data products from a notebook prototype to a deployed, observable service — covering model training, inference serving, ingest pipelines, CI/CD, and monitoring.
Currently building CHRONO, a retail-facing quantitative trading platform that aims to bring hedge-fund-grade analytics within reach: a Transformer-based regime forecaster, a 22-signal alpha library, and a Markowitz + VaR risk engine — all wrapped in a real-time, observable stack.
What I optimize for: production reliability · data correctness · observable systems · honest metrics over flashy demos.
| Project | Stack | Highlights |
|---|---|---|
| CHRONO private | FastAPI · Next.js 16 · PostgreSQL / TimescaleDB · ONNX Runtime · Docker · GitHub Actions | Transformer v2.3 (56.9% OOS) · 22 alpha signals · 9-state regime detector · shock detector · self-correction loop · real-time WS ticks |
| PII-Guardian private | Python · HyperCLOVA · NCP · Streamlit · CI/CD | LLM-driven PII leak detection deployed on Naver Cloud — full pipeline from log ingest → model inference → alerting |
| im-bank-n8n-agent | n8n · Node.js · Upstage Solar AI | Real-time security log analyzer & auto-learning agent for IM Bank — workflow-driven LLM ops |
| Lecture-Summarizer-AI private | Python · OpenAI Whisper · Streamlit | LMS 강의 실시간 녹음 + Whisper STT + LLM 요약 파이프라인 |
| opengl-earthquake-simulation | C++ · OpenGL | 3D 지진 대피 교육 시뮬레이터 — 인터랙티브 카메라 + 시네마틱 가이드 뷰 |
| battle-rogue | Unreal Engine 5 · Blueprints · Dedicated Server | 1:1 온라인 대전 격투 게임 — UE5 데디케이티드 서버 |
| Languages |
|
| Backend / Data |
|
| ML / Inference |
|
| Frontend |
|
| Infra / DevOps |
|
| Observability |
|
- Boundaries over abstractions — validate at I/O edges, trust the core. Don't add error handling for things that can't happen.
- Reproducibility first — every model, every dataset, every deploy is versioned and traceable.
- Observability before scale — no logs / metrics / traces, no production.
- Automate the boring guardrails — pre-commit hooks, CI gates, security scans, dependency review.
- Honest metrics > pretty demos — OOS over IS, ablations over cherry-picked screenshots.
대구대학교 컴퓨터소프트웨어 · ML / Data Systems · Daegu, South Korea
README · stats · snake animation are auto-refreshed daily via GitHub Actions.