I began my career in architectural design. Now I build AI products.
The two fields may look unrelated, but they form a continuous path for me. The way I once structured spatial constraints and circulation now shapes how I design data flows and AI systems. Architecture also taught me that expert work resides not in documents alone, but in constraints and exceptions, review criteria, and judgment.
That is the pattern behind most of my work today. I observe ambiguous real-world processes to identify the right problem, translate experts' tacit knowledge into prompts, business logic, evaluation criteria, and data contracts, then build products that people actually use—from generation and validation through storage and user selection.
I do not choose the technology first. I first ask how often the problem recurs, what impact a solution could have, and how much validation will cost.
Experience
aa-architecture · Head of AI Lab / DX·AX Lead
Jun 2026 – Present (Full-time)
As the sole in-house developer and first DX/AX lead at an architectural firm of ~100 staff, I oversee the technical roadmap, product engineering, cloud infrastructure, and company-wide adoption.
- Driving a modular human-in-the-loop schematic design automation pipeline:
site planning → modeling → visualization → presentation - Developed beta tools for map-based building ledger lookup and DXF/SketchUp MCP modeling automation
- Built an image generation and LLM/VLM evaluation assistant agent, capturing user feedback (selection, regeneration, ratings) for iterative improvement
- Built GCP infrastructure (Cloud Run, Cloud SQL, GCS, Secret Manager) with Terraform and automated keyless CI/CD using GitHub Actions & WIF/OIDC
- Led Microsoft 365 deployment with a 15% negotiated cost reduction; established structured data accumulation for future internal RAG systems
- View Project Details →
Bookips · AI Engineer
Jun 2025 – May 2026 (Full-time)
Designed generation and validation logic for CSAT English question pipelines, focusing on cost and quality optimization.
- Structured expert tacit knowledge into LLMs and deterministic business logic, raising initial internal review pass rate from under 10% to ~70%
- Maintained 0 quality complaints over 6 months and improved answer generation PoC accuracy from 91.21% to 96.45%
- Expanded the valid question pool from ~40k to 90k using integer programming and greedy algorithms without additional LLM generation costs
- Validated low-cost models via Log Probability: achieved 100% agreement with baseline while reducing verification cost by 90% (KRW 50 → 5) and latency by 50% (30s → 15s)
- Question Generation & Validation → · Answer-Choice Optimization → · Cost Optimization PoC →
Drift Patent · AI Engineer
Feb 2025 – Mar 2025 (Contract)
Solely developed the production patent specification drafting pipeline for an IP tech startup.
- Implemented a LangGraph multi-agent generation-validation-revision workflow with patent drawing image input
- Built robust error handling (token cutoffs, prompt injection defense, auto-revision) and delivered production Docker containers
- Improved quality score against patent attorney review criteria from 20–30% to 80–90%
- View Project Details →
Education
Fastcampus Upstage AI LAB
Jul 2024 – Feb 2025
AI Engineering Intensive Program (Machine Learning · NLP · RAG · MLOps)
- 2x 1st Place, 1x 2nd Place across 4 competitions (e.g. 2nd place in RAG Document Retrieval: View Project)
Queen’s University Belfast
Sep 2015 – Jun 2019
BSc Architecture (Belfast, Northern Ireland, UK)
Skills
- AI / LLM: LangGraph · LangChain · RAG · Prompt Engineering · LLM/VLM Evaluation & Serving · Log Probability Validation
- Backend / Data: Python · FastAPI · PostgreSQL · SQL · Prefect · dbt · Docker · MLOps · Batch Pipelines
- Cloud / Infra: GCP (Cloud Run · Cloud SQL · GCS · Secret Manager) · Terraform · GitHub Actions (WIF/OIDC Keyless CI/CD)
- Domain: DXF · SketchUp MCP Modeling Automation