Microsoft Research Lab – Africa, Nairobi

January Seminar

GeoMind: A Multi-Agent Framework for Geospatial Decision Support

Join us on Thursday January 29th at 3pm EAT for a talk by Muhammad Sohail Danish (opens in new tab)

Rapid access to actionable geospatial insights is essential during disasters such as floods, wildfires, or earthquakes, where timely decisions can save lives and resources. In many scenarios, especially in low-resource settings or when GIS experts are not immediately available, policymakers, humanitarian responders, and other non-technical users need to quickly understand the impact of events, such as identifying damaged buildings or locating the nearest hospitals to a damage zone, without navigating complex GIS workflows or switching between multiple tools. While recent efforts to integrate natural language interfaces with GIS systems show promise, existing solutions remain limited: they often support only simple queries, lack multi-step spatial reasoning, and struggle with complex spatial joins across heterogeneous datasets. Furthermore, Large Language Models (LLMs) can exhibit hallucinations when reasoning over multimodal geospatial data, making them unreliable for high-stakes disaster scenarios.

To address these challenges, we propose GeoMind, a multi-agent system designed to bridge natural language understanding and advanced geospatial analytics. GeoMind enables non-experts to query multi-layered geospatial datasets using natural language and receive accurate, context-aware insights to support critical decision-making during emergencies.