AI/ML Platform Engineer · LLM/RAG · Data Platform
Loaded and processed roughly 300M Korea Customs Service and MFDS records from 2022-2026, then built admin features, automated DB refresh, operations handover, and load-test workflows. The architecture is refined around development, operations, and user workflows.
Structured ASR, LLM, RAG, ffmpeg, auth/admin/API, and worker components into 5 Compose services. Output quality is managed through 29 API routes and 225 test functions.
A market-data GraphRAG system with 31 FastAPI routes, 78 DB tables, 12 hypertables, and 67 indexes. Collection, retrieval, analysis, backtests, and risk checks are connected in one operating surface.
Integrates and analyzes highly unstructured source documents to auto-fill public customs import declarations.
Runs PCAP → features → classification → evidence, similar cases, summary → report in one flow. Detection trust is improved with probability calibration, Unknown routing, SHAP, and FAISS.
babayetu <alias> one-command access.
RSA auto-login, native backup/restore, and Windows Terminal integration
improve both security and recovery.
English: professional working communication
I design from data ingestion to deployment and reporting with a product operations mindset.
LGBM, SHAP, Unknown routing, RAG, LLM, probability calibration, and evidence design.
Learn More →Batch jobs, ETL, OCR preprocessing, caching, fallbacks, lazy loading, and operations-friendly design.
See Resume →APIs, Gradio UI, PDF reports, monitoring, and alerts.
Case Studies →Admin APIs, authentication, workers, operating screens, and deployment documentation.
Additional Work →Additional work that shows implementation range beyond the main target roles.
Public case evidence: roughly 300M Korea Customs Service and MFDS rows loaded and processed from 2022-2026, plus DB migration, automated refresh, and load testing.
View evidence →31 FastAPI routes, 78 DB tables, 29 video API routes, and 225 test functions.
View case study →