Dekart

Claude Can Now Analyze Your Fleet Data and Create Maps

by Vladi ‐ 2 min read

Turn Claude into a fleet analyst with the GeoSQL skill.

Fleet manager job is hard. You must be product, data analyst, scientist and procurement specialist at the same time.

Fleet dashboards lag behind real cases on the ground. I often see fleet managers doing their own math in spreadsheets 🤯

Being able to use Claude for ad-hoc fleet analytics may save your day!

But Claude alone, even with access to a semantic layer, cannot do fleet analytics, for the same reason a human can’t do it by looking at a table with coordinates.

I’m using Dekart and the GeoSQL skill for Claude, that solves it by adding a “map harness” to the agent. Instead of “SQL and Pray”, Claude loops on SQL → Map until it achieves plausible results 🧠

The map

Recommended rebalancing plan: short hops from the surplus ring into the downtown deficit core.
Recommended rebalancing plan: short hops from the surplus ring into the downtown deficit core.

View interactive map

In this map I connected my data warehouse account to Dekart and installed the GeoSQL skill, then asked Claude “where to move bikes before the evening peak”.

Claude analyzed 592,555 trips / 832 stations, detected evening peak hours (16:00–19:00) and computed net flow: 336 stations drain, 447 accumulate. Then it created a recommended rebalancing plan and built the map 🤩

Best part, the map is always up to date, connected to my data, I can use it again and share with a team! Dekart is open-source, supports BigQuery, Snowflake and Postgres, can run locally or self-hosted.

Map in the loop

GeoSQL is not just a markdown file. It’s a map-in-the-loop Claude harness, running inside Docker.

GeoSQL’s main goal: turn Claude into a competent fleet analyst that frees your time from routine and ad-hoc requests. It removes the main blocker, Claude hallucinations on geospatial, by adding a map-in-the-loop flow.

Claude writes the SQL, Dekart runs it and renders the map, and Claude sees the map.
Claude writes the SQL, Dekart runs it and renders the map, and Claude sees the map.

It has three components:

  • GeoSQL Skill: markdown files.
  • Dekart CLI: interface for Claude to query your geospatial database and render the map.
  • Dekart: the Docker application based on KeplerGL, with connectors to PostGIS, BigQuery, and other GIS SQL databases.

To install it you need Python + run one command:

pip install geosql && geosql

Learn more on GitHub: https://github.com/dekart-xyz/geosql