Data Analytics Engineer building AI-powered automation systems and data pipelines.
Currently working on AI-driven process automation - OCR, RAG pipelines, and LLM integrations that cut manual work by 90%+.
- Data & Engineering: Python, SQL, R, Power BI, Process Mining
- AI / LLM: RAG pipelines, LangChain, OpenAI/Claude APIs, ChromaDB
- Automation: Power Automate, RPA, OCR, ETL pipelines
- Tools: AWS, Git, Java, HTML
- AI Expense Automation — OCR + LLM system that reduced reconciliation time by 90%+ (Accuver America)
- RAG Competitor Intelligence — Document analysis pipeline extracting insights from unstructured data
- Sustainability Data Audit — Python + Power BI system that found $20K+ in cost discrepancies (Andersen Corp)
- Process Mining Analysis — Analyzed 3M+ records to improve insurance conversion by 5% (PuzzleData)
- Walmart Sales Forecasting — Time-series forecasting & ML model comparison
- Process Mining Portfolio — Procurement workflow optimization (~80%+ conformance)
- 보다 (Boda) — Vision OCR (Gemini to Claude) that turns receipt images into structured expense data + Excel
Open-source plugins & skills I build for Claude Code:
Parallel workflow
- ddaro — Worktree-based parallel workflow: isolated branches, CI-orchestrated merges, crash recovery.
- pumasi — Claude as PM, Codex CLI as parallel developers — orchestrated parallel coding.
Multi-model review
- prism — Multi-angle code review: 5 parallel agents + a singleton verifier pass.
- triad — 3-perspective deliberation (clarity / longevity / comprehension) until consensus.
- mangchi — Cross-model code hardening: Claude writes, Codex CLI critiques one axis at a time.
- prism-devil — Aggressive red-team probe: single-agent attacker-mindset review with auto-loaded checklists.
Deep research as verified RAG
- batchim (받침) — Verification-gated deep research: isolated-verifier + N=3 panel entailment, code-enforced span/number anchors, and an auditable signed ledger — RAG you can trust.
Knowledge distillation
- galmuri (갈무리) — Distills reusable, project-agnostic techniques from a project into portable technique cards.
- knowledge — The card library galmuri writes to: self-contained, reusable engineering technique cards.
I treat AI coding as an engineering system, not a single chat:
- Parallel, isolated branches — worktrees keep concurrent work from colliding (ddaro).
- Cross-model review gates — Claude and Codex critique each other before anything merges (prism / triad / mangchi).
- Verification before trust — research and retrieval must cite, anchor, and pass entailment checks, not just sound right (batchim).
- Distill what works — recurring solutions become reusable technique cards (galmuri → knowledge).
- Email: minwoo.park219@gmail.com
- LinkedIn: linkedin.com/in/mp74484
- Portfolio: minwoopark.dev

