Teacher uses Claude Code and Markdown for home management

Scott Farrar, a math teacher and parent, built a household management system using Claude Code to read and write Markdown files, creating a transparent, persistent memory layer for AI interactions. Unlike standard chat interfaces where context is opaque, this system stores specific household data—including appliance serial numbers, warranty expiration dates, insurance quotes, and repair histories—in local files. Farrar uses this setup to handle complex tasks like comparing car insurance policies, where the AI tracks quotes, identifies coverage gaps, and generates checklists for next steps, such as emailing specific agents. The system also manages a comprehensive to-do list with priority levels and cost estimates for repairs, including a multi-thousand dollar dry rot issue.

For collaboration with his wife, Farrar uses a hybrid approach of Dropbox for general files and GitHub for scripts that lint the data and export to a DACboard wall display, which surfaces top-priority tasks and recent wins. The setup allows Farrar to ask loose questions like 'do I need to plan for earthquakes?' and receive grounded advice based on his specific home context, such as foundation type and liquefaction zone status. Farrar argues that using coding agents with local file systems provides greater confidence and control over AI advice compared to opaque web-based chatbots. "I wanted to build a system where I was confident that AI was grounded in my truth," Farrar said.