Natural-Language Geospatial Query Platform

Architecture

1. Question

   plain words
2. Language model

   writes the query
3. Checks

   only the data granted
4. Spatial query
5. Answer

   text · chart · map

What it does

Public geographic and statistical data is rich, and nearly unusable for anyone who can't write database queries across many sources, each with its own jurisdictions and geometries. The hard part of answering questions over data like this is not the language model. It is describing the data so the model can reason about it, and keeping every answer inside the data that user is allowed to see.

I co-founded the company and built its first versions. I originated the query system, where a question becomes a spatial query that is checked and kept within the data sources the conversation was granted, then runs and comes back as an answer, a chart or a map. I also originated the user interface and the payment experience and its integration, and set up the infrastructure it runs on.