Work in data analytics and applied AI.
Usually starts with "why does this number look wrong?" — then pulling apart the process, redefining the metric, building the system. Habit of talking to people before touching the data.
Worked on-site in Malaysia, Singapore, Thailand, Brazil, and the US — learned firsthand why the same metric produces different numbers across countries, and spent most of that time standardizing data and building KPI frameworks from scratch.
These days mostly thinking about behavioral analytics pipelines and how to build AI systems that actually hold up in production.
** material-category-mapping-ai** Auto-classification pipeline for ~100K multilingual material records. 80%+ accuracy within a standardized category schema. Human-in-the-loop design — accuracy compounds as users give feedback. Triplet Loss + Hard Negative Mining based training architecture.
** vendor-deduplication-ai** Dedup pipeline across ~50K global supplier records. No universal ID in overseas data, so tax ID used only as secondary signal. 3-stage hybrid: Blocking → ANN → Embedding similarity scoring. ~30% duplicate rate found across the dataset — data cleansing completed.
** Internal Policy Review RAG System** (internal project) Automated compliance and subcontracting law review system. LangChain + LLM + internal docs / external legal data. Local LLM setup using Ollama / DeepSeek.
** phishing-detection** AI proxy that answers unknown calls and detects scams in real time. Whisper STT + keyword scoring + GPT response strategy. Flask backend · GitHub Actions CI/CD · deployed on Render.
** kleague-analytics** Pass destination prediction from K League event sequences. LSTM vs Transformer comparison. Sequential behavior modeling.
Analytics / ML
Python R SQL pandas scikit-learn PyTorch TensorFlow
NLP / Embeddings Triplet Loss LangChain RAG Ollama Streamlit
Systems / Backend
Flask Spring Vue.js Oracle DB
GitHub Actions Render UiPath RPA
ERP / Enterprise
SAP ERP SAP Ariba SAP SRM SAP BW SAP SAC
CJ Cheiljedang — Procurement Strategy / Next ERP TF (Aug 2022 – present) Global SRM PI, procurement dashboard, AI material classification, vendor dedup, RAG system
Hansol PNS — Digital Innovation (Dec 2018 – Aug 2022) 20+ AI / automation projects across manufacturing, logistics, pharma