Using AI to collect data

I ran a livestream working through the first phase of a multi-part project: recreating my Oregon by the Numbers report for other states, starting with getting the data. I used Claude Code inside Positron to write the import scripts, troubleshoot renv, and set up the repo.

You'll learn:

• How to prompt Claude Code to write tidycensus and tigris data import scripts organized into clearly labeled sections
• How Claude Code's plan mode lets you review what it intends to do before it touches your files
• A way to use Claude Code to write and commit git messages in logical chunks rather than one big commit
• How to copy globally installed Claude skills into a project repo so collaborators get access to them
• How to set up a Claude hook in settings.json so the README is automatically checked and updated after every session

Tools mentioned: Claude Code, tidycensus, tigris, renv, Positron, GitHub Desktop, Raycast, Solo, Quarto, Typst, ggplot2, tidyverse

Key moments:

• 00:00 Stream starts, David checks audio and chat
• 01:03 Overview of the three-part livestream plan
• 02:26 Introduction to the Oregon by the Numbers report
• 05:00 Creating a new repo from a GitHub template
• 07:03 Opening the cloned project in Positron
• 09:06 Using Claude Code extension inside Positron
• 22:07 Prompting Claude to write the tidycensus import script
• 30:52 Reviewing the generated import-data.R file
• 43:11 Adding shapefile imports with the tigris package
• 46:42 Quick map check to confirm county boundaries imported correctly
• 47:12 Asking Claude to write and push commits in logical pieces
• 49:21 Switching to Solo to copy global Claude skills into the project
• 57:32 Asking Claude to update the README and keep it current via a hook
• 01:03:44 Hook and settings.json explained, final commit and push