AI

[Agent] korea-persona-interview, a Korean Synthetic Persona Interview Tool Powered by CLI and MCP

28 minute read

A writeup of korea-persona-interview, a side project that automates Korean synthetic persona interviews. It layers multi-turn interviews, automatic follow-ups, and persona drift detection on top of NVIDIA’s Nemotron-Personas-Korea dataset to validate business hypotheses quickly. The post also covers the design pattern that lets three entry points (CLI, MCP server, and MCP orchestrator) share a single core.

[Agent] Yangsiljang, a Real Estate Law & Precedent RAG Chatbot Built as a Side Project

21 minute read

A breakdown of the architecture and implementation details behind Yangsiljang, a real estate law and precedent RAG chatbot built as a side project. The frontend was built with Claude Code’s help; the backend was written from scratch. Topics covered include a custom Korean tokenizer build for Milvus, Qwen3 embedding/reranker, HyDE, SSE streaming, and data ingestion from the Korean Legislative Service.

[cmux] A Ghostty-Powered Native macOS Terminal for Parallel AI Agents

4 minute read

cmux is a native Swift + AppKit macOS terminal that embeds the Ghostty engine and is built specifically for running multiple agentic coding assistants like Claude Code, Aider, and Cline side by side. This post covers installation, core features, and keyboard shortcuts.

[Claude Code] Building a Real-Time Agent Dashboard

23 minute read

How I built a real-time monitoring dashboard to visualize what 19 Claude Code subagents are doing - from a failed Gather Town experiment to a practical WebSocket-powered control center.

[Kiro] Agent Configuration

3 minute read

A guide covering agent creation, configuration, prompt modularization, and sub-agent usage in Kiro.

[LLM] Setting Up Ollama Remote Access

1 minute read

Configuring the OLLAMA_HOST environment variable and Windows Firewall to access a local LLM from other devices on the same network.