Presentation in preparation...
From Language to Action: Transforming Geospatial Data with AI
10/01/2025 | 14h00 -> 14h40 | Room A Forum: DATA / ANALYTICS / AI / PRIVACYWhat if asking a question about a place were as simple as saying it in plain language?
Today, large language models (LLMs) do more than just understand words — they can search for geographic data, structure it, link it together, query it, and even control tools to produce analyses and maps.
In the geospatial field — where data is often scattered, complex, and locked in technical formats — this ability to bridge human language and digital geoinfrastructure is a quiet revolution. But a decisive one.
It’s no longer just about “finding the right layer” in a data warehouse — it’s about answering real questions about places.
What are the climate risks in a given area?
What’s the best soft mobility option between two towns?
How does a local urban plan interact with topographical, environmental, or social data?
We’ll show how LLMs go beyond simple retrieval, acting as geo-intelligent agents that can:
- converse with catalogs and knowledge graphs,
- interpret user intent,
- generate queries on data platforms,
- and control tools like QGIS or the MCP protocol to automate analytic and cartographic tasks.
In short, the LLM becomes a smart, flexible interface between humans and the entire geospatial ecosystem.
We’ll share insights from several real-world projects where this approach is taking shape:
With swisstopo, SchweizMobil, and SuisseRando, we’re testing AI-assisted search and storytelling to bring cartographic data to life.
With DataGrandEst and Géo2France, we’re experimenting with chatbots that directly query regional datasets.
And as part of the Digital Twin of French Territory (IGN, CEREMA, Inria), we’re exploring how to connect models with a nationwide knowledge graph.
What we’re proposing is a paradigm shift:
No longer just consulting data - but conversing with it.
No longer just using tools - but activating them through conversational intelligence.
A territory is a collection of data but it’s also a collection of questions. And maybe, tomorrow, we’ll be able to talk to them.