If you check into the docs, skill is the entry point for the map harness for Claude. You need to install the skill and run the Docker image.

That said, some of the skill frameworks like gstack created 10x productivity gain for me. IMO, worth sharing here.

Map snapshot PNG. Apparently, LLM is quite competent when reading map images. It can say, “Oh, that's not all London coverage.” “ “Oh, this and this street is a problem (without having street data).”

Didn't expect a plain screenshot to be enough for that. Thanks.

Plus one. (I’m the author of GeoSQL.) This is why I personally store data in local PostGIS. The whole map harness is running locally, except for Claude. I did not write SQL since April. I am making 1-2 analytics projects per week.

Here is a video explaining roughly how I work now: https://www.youtube.com/watch?v=JCOhkE0rPWA

It's not % it's number of evals. Here is how I did evaluation https://dekart.xyz/blog/how-to-evaluate-claude-skill-output-...

Hey, I'm the author of GeoSQL. Found this thread after realizing the repo got 400 stars last week. Ask me anything.

Thanks for sharing, and thanks for making the video you shared. A couple thoughts. Is PostGIS king in this area? I've been liking duckDB as it does not require a server. Is there a good alternative to PostGIS in the duckDB world? Maybe just plain GeoParquet files read in through duckDB?

In the example in your video are any special GEOMETRY functions being used in the underlying SQL? Or, could your data just have been in plain postgres?

A similar product GeoSQL is Malloy which puts a semantic layer on top of your data for better LLM understanding. Malloyyo gives you an MCP server for precise and auditable interaction with your data.

Hey HN,

I’ve been working on a backend for Kepler.gl and testing it with public datasets from BigQuery. I thought it might be interesting for the HN audience to see what kind of free spatial datasets exist in BigQuery marketplace and how they can be visualised.

Share more datasets and maps in the comments!