I'm a builder based in Seoul, working at the intersection of:

• Agent systems for planning, execution, and review
• LLM infrastructure for real-world products
• Document / OCR / parsing pipelines that turn messy inputs into structured data
• Developer tools that make AI systems easier to evaluate, operate, and improve

My recent work is centered around LangGraph, LangChain, FastAPI, PostgreSQL, Docker, and pragmatic Python architecture.

I design and experiment with stateful agents, coding agents, and supervisor-style workflows.

Key interests:

• structured output first design
• tool schema design with Pydantic
• planner / implementer / reviewer loops
• memory and state synchronization
• balancing ReAct loops vs structured orchestration

I build developer-facing AI products that solve practical problems, not demo-only toys.

Areas I care about:

• safe LLM integration
• PII protection and redaction pipelines
• evaluation-aware product design
• retrieval and transformation pipelines
• robust backend architecture for AI services

I'm interested in systems that convert unstructured sources like PDFs, markdown, HTML, OCR output, and blog posts into clean, indexed, structured data.

I care a lot about:

• Domain-driven design
• Dependency inversion
• Injectable abstractions
• Async Python
• Clear system boundaries
• Build first, then rewrite cleaner

I like systems that are simple to operate, easy to test, and honest about trade-offs.

Languages
Python C++ Kotlin

AI / Backend
LangGraph LangChain FastAPI Pydantic PostgreSQL Docker

Infra / Tools
Qdrant OpenAI-compatible APIs CachyOS fish PDM

Before going deep into AI systems, I worked on:

• custom game engine development in C++ / OpenGL
• mobile app development with Kotlin
• architecture-heavy application design across multiple domains

That background still shapes how I think: I care about performance, architecture, boundaries, and long-term maintainability.

Right now, I'm especially interested in:

• building better custom agents
• evaluator and benchmark driven development
• document normalization pipelines
• practical LLM product architecture
• local + hosted hybrid AI workflows

• Email: miracle.cowf@gmail.com
• GitHub: MIRACLE-cowf