I like deep neural nets. - AI Engineer with 4+ years of specialized AI/ML experience and 19+ years of IT infrastructure background.
I design and build RAG systems, LLM applications, and end-to-end AI pipelines — from data collection and model training to cloud deployment and monitoring.

• 🏆 1st place — Upstage AI OCR Competition (2025.12)
• 📄 SCOPUS-indexed international journal publication — GCN-based Fire Situation Recognition (2024)
• 🏛 Led government R&D AI projects (ETRI, IITP) as AI Project Lead
• 🤖 Built RAG/NLP systems deployed in real-world services (112/119 emergency inference, B2B matching platform)
• ☁️ Experienced in MLOps: MLflow, GitHub Actions, AWS EC2/S3, Docker-based CI/CD/CT pipelines

LLM · RAG

ML · DL

MLOps · Cloud

Backend · Infra

• Researched and developed a fire-situation recognition model based on a Graph Convolutional Network (GCN)
• 95% prediction accuracy achieved · Published in a SCOPUS-indexed international journal
• AI Project Lead · Owned the full pipeline: data collection, preprocessing, labeling, and model training

• Designed a receipt text-detection model and optimized polygon-coordinate detection
• Built a robust detection pipeline across 3,600+ images
• Maximized generalization performance through data augmentation and hyperparameter tuning

• Developed an NLP model and API to infer urgency level (Code 0–4) and incident type from emergency-call text
• AI Project Lead · Led a team of 3 data scientists and 2 AI developers

• Converted a MATLAB-based signal-processing application developed at ETRI (Electronics and Telecommunications Research Institute) into C++ software for a Windows desktop application
• AI Project Lead · Led a team of 1 data scientist and 3 AI developers

• Designed an ETL pipeline for 400GB of Russian patent data → produced 30,000 high-quality JSON records
• Developed a Multilingual BERT-based enterprise-matching classification model · 90% classification accuracy
• Devised a three-stage cross-mapping table (IPC ↔ KSIC) and built a VPN-based crawling engine to collect a database of 100,000 Russian companies

• CCTV video-based missing-child detection project using a YOLO model, conducted at ETRI (Electronics and Telecommunications Research Institute)
• AI Project Lead · Led a team of 3 data scientists and 2 AI developers

• Built an end-to-end RAG service: constructed a ChromaDB/FAISS vector database from self-scraped LangChain technical documentation
• Improved Top-k retrieval accuracy by 30% · Docker Compose CI/CD · Implemented a streaming UI

• Combined Ministry of Land, Infrastructure and Transport transaction data with subway/bus accessibility data through feature engineering
• Built an XGBoost/LightGBM ensemble model · Reduced RMSE by 35% versus the baseline

• 📧 sizin@naver.com
• 💼 LinkedIn
• 🐙 GitHub