Build a live traffic and flood map in a day from open data
A Thai tech founder told a Claude meetup he used Claude Code to build a 3D Bangkok map of traffic and flooding within 24 hours, without Google's map API, using a second AI to check the work.
What they did
The build pulled OpenStreetMap, Open-Meteo and Thaiwater data, and used computer use to read hard-to-parse weather sites and turn them into plain summaries. He set up a workflow with separate build and validation steps, reviewed the architecture decision records himself, and had Claude write progress to a file and clear its context at around 60% full. Claude Code did most of the coding, Codex handled the 3D parts, and another model reviewed the output, with an AGENTS.md file so every tool shared the same project structure.
What happened
The speaker says the working map was done within the 24-hour challenge, a job he contrasts with how hard a flood site was to build years ago.
Keep in mind
This was a personal challenge project described at a meetup, not a commercial deployment, and it is reported through an attendee's recap. Data accuracy and upkeep are not discussed.
Try it with your agent
Build a web map of <city> that shows <first data layer> and <second data layer> using only open data sources such as <data sources>. First write a short plan and architecture decision record for me to approve. Then build in small steps, run checks after each one, and when your context gets long, write progress to PROGRESS.md before continuing.
Replace the <angle brackets> with your own details.