AI Agent Use Cases
Views —

AI Agent Use Cases

Added today

10All 221

1–20 of 221

Dev

An agent team that builds a map showing the causes of Singapore's haze

The author used Claude Code to build a haze map displaying smoke from Indonesian fires and wind on one screen.

Claude Code

How it works

  1. The author decided what to build and which claims the map could support. Claude Code wrote the data-processing code, map, particle model, and CI.
  2. Multiple Claude Code sessions ran in separate git worktrees. Every change began with a GitHub issue, became a PR with before-and-after screenshots, and was reviewed and merged by a lead session.
  3. Rules were created to filter false positives in satellite fire data, including gas flares, factories, volcanoes, and reflections over water.

Results

The map was released after removing 51,922 false positives. According to the author, GitHub Actions now refreshes the data twice a day, and a Cloudflare Worker alerts them when it becomes stale.

Notes

The map is free and does not require login. Forecasts and projected PSI ranges are labeled as forecasts. Development time and cost were not disclosed.

Source

Reddit Original post ↗

Example prompt · Claude Code

Build a web map showing where air pollution in [region] may be coming from on one screen.

Inputs: displayed region [region], time period [starting date], public datasets [e.g., satellite fire detection, air-quality indexes, wind forecasts], and interface language [language]. First check each dataset's terms of use, update frequency, attribution requirements, and whether it is free; show the findings in a table.

Implementation: 1) Write a Python script to fetch and clean the data. 2) Filter false positives such as factories, volcanoes, and reflections over water from satellite fire data using explicit rules, recording the number removed by each rule. 3) Overlay fire locations, smoke, and wind by date on the map. Always label forecast data as forecasts. 4) Organize changes into issues and PRs, attach before-and-after screenshots, and merge only after my review. 5) Add a scheduled data-refresh job and a check that alerts me when data becomes stale.

Deliverables: a mobile-accessible map URL, a list of data sources, false-positive filtering rules and counts, and one paragraph of operating instructions. First build the map with one day's data and let me check for remaining false positives before enabling the full period and automatic refreshes. Review the wording so the map does not assert unconfirmed causes.