I build data + AI products end-to-end — from unstructured data pipelines to LLM-powered systems running in production — and I care most when that work connects to real business impact.

• Started in NLP data analysis, expanded into data engineering and AI agent development.
• Comfortable across the whole stack: data collection → ETL → analysis → datamart → product.
• Currently deployed as a Forward Deployed Engineer (FDE) at a public institution, building production AI agents.

LLM / AI Agent

Backend / Data

Cloud

Distributed / Big Data

Multimodal LLM Influencer Analysis & Data Productization — WHOTAG Pipeline that analyzes global influencers (image / video / caption) with multimodal LLMs and turns the results into data products.

• Designed the analysis → datamart pipeline; modeled influence / authenticity scores via 1-vs-1 win-rate comparison.
• Cut analysis cost ~90% by moving repeated key-based lookups from BigQuery full scans to Firestore.
• Shipped the output as Data API / MCP / GPT Apps. Vertex AI Firestore BigQuery Python

Natural-Language Influencer Search SaaS — WHOTAG (Global B2C) Search engine that finds influencers from plain-language queries across 2.4M+ creators / 120 countries.

• LangGraph intent routing + text2sql / vector hybrid search (vector fallback when SQL fails).
• Trimmed storage cost 84% by pruning unused Firestore indexes based on real call statistics.
• Launched globally — peak MAU 10K, up to 10K queries/day at ~sub-cent per query. LangChain LangGraph text2sql Qdrant Docker CI/CD

Agentic Weekly Movie-Review Analytics — CJ CGV Automated system delivering AI-analyzed reviews of each week's new releases to a major cinema chain.

• 9-category classification + unsupervised clustering (HDBSCAN / UMAP) for sub-topic discovery.
• Auto-generated reports (Markdown → Marp), with edits delegated back to the LLM.
• Agentized weekly operation with Claude Agent SDK — ~90% less operating effort; PoC → annual contract. Claude Agent SDK LangGraph Python GCP

Social Big-Data NLP Analytics Engine — Infrastructure The in-house engine that powers B2B social-data reports: ETL, document management, and analysis indexing.

• Took over and ran a 40-node Hadoop / Spark indexing pipeline with zero downtime (tens of GB/day).
• Built hash-based deterministic sampling — same sample reproduces even as the document pool changes.
• Stood up collection monitoring + auto-recovery; migrated the engine to AWS with lifecycle-based cost control. Hadoop Spark Airflow dbt AWS (Glue / Athena / EMR)

• Building production AI agents as an FDE at a public institution.
• Interested in agent orchestration, retrieval, and data infrastructure that holds up at scale.

• Co-author, "Korean adolescents' coping strategies on self-harm, ADHD, insomnia during COVID-19: text mining of social media big data" — Frontiers in Psychiatry (SCIE, 2023). Text-mining of ~12.5M adolescent social posts.