When a flood crests or a fire perimeter shifts, the emergency managers who need spatial answers fastest are often the least likely to open a GIS or write a spatial query. Their critical, time-constrained needs often end up in the ticket queue awaiting a dedicated GIS analyst's support. The data exists: flood zones, wildfire susceptibility, locations of critical facilities, and more. However the need for deep technical know-how often sits between spatial data and an actionable result.

NV5 would like to present GeoIQ, a system that lets anyone "Ask the Map" in natural language. GeoIQ's goal is to empower subject matter experts to work in GIS environments, allowing them to execute basic GIS operations through plain language. For experienced GIS power users, GeoIQ provides a much more ergonomic interface to work with large, complex GIS data.

GeoIQ is built on OSGeo/FOSS4G tools: QGIS and OpenLayers provide the front-end integrations, while GeoIQ’s backend is built on GDAL/OGR, GEOS, PROJ, and PostGIS. By embracing OSGeo/FOSS4G’s open standards GeoIQ feels native to users, living in their preferred desktop and web environments, and using the same set of open tools. The result is that GIS analysts are more “in the loop” than ever, with tools to audit their AI outputs. By leveraging these existing widely used GIS tools, GeoIQ brings the "human-in-the-loop" forcefully, which is critical for ensuring an AI's responses are accurate and precise. GIS tooling provide a wealth of visualization and measurement capabilities for validating an AI's results in near real-time, as well as extending the capabilities of off-the-shelf LLM models for spatial analysis.

This talk will begin with a live demo to introduce GeoIQ and showcase natural language Q&A in an open source GIS environment. The demonstration will emphasize disaster management workflows. Afterwards, we will feature a short discussion of GeoIQ's technical design and approach to AI safety. We'll be candid about the stakes. An AI that confidently gives a wrong information during a disaster is worse than no AI at all, so we will conclude on techniques for safely developing, shipping and monitoring spatial AI.