Blog
OCR + Rule + LLM
영수증 사진을 올리면 OCR이 읽고, Rule이 좁히고, LLM이 뽑아낸다 — 근거 없는 값은 절대 반환하지 않는다
Built a local-first receipt understanding pipeline that separates concerns across three explicit stages: OCR extracts text, Rules narrow candidates, and LLM performs semantic extraction — each doing only its own job.
The key principle: no hallucinated evidence. The LLM returns a value only when it …
AIRead more →
RAG Chatbot with Multi-LLM Com
PDF를 올리면 AI가 증거를 찾고, 두 가지 전략으로 답한다
Built a transparent RAG chatbot that makes the full pipeline visible — retrieved chunks, merged context block, and prompt text are all shown in the UI, not hidden.
The same question is answered two ways simultaneously: STRICT returns answers only when evidence explicitly supports them. …
Read more →AI Agent with Gemini & Slack
이메일이 오면 AI가 읽고, 판단하고, Slack으로 알린다 — 사람 손 없이.
Built an agentic workflow that classifies incoming emails by urgency using Gemini, then automatically sends the result to Slack. The same 3-node pipeline was implemented three ways — n8n, LangChain, and LangGraph — to directly compare how each tool expresses orchestration.
The core architecture: Orchestration …
AgentRead more →
Claude Cowork: Automation
Cowork으로 만드는 나만의 자동화 루틴
Building My Personal Automation Routine with Cowork
I've always struggled with the same two problems. My disk fills up silently until something breaks. And my mornings start scattered — no clear picture of what's happening in AI, what I need to study, or what's on …
AIRead more →
Building AI Agents
AI Agent 구현해보기
Building Agentic Flows: A Hands-On Project Plan
I've been putting together a plan to build and teach a series of hands-on Agentic Flow projects. As someone who's been working across software development, AI research, and QA automation, I've always wanted to bridge the gap between …
AI AgentRead more →