We'll discuss recent advancements in fAIr. Specifically, we'll explore our approach to enabling developers to share their GeoAI models for humanitarian mapping purposes. We'll also dive into the development of our MLOps architecture, which leverages STAC, MLM, Zenml, and Kubernetes to run models independently in the cloud.

We share how the Open Exposure Taxonomy integrates several open data taxonomies and improves the quality of data for disaster risk.

Using cheap 250 g consumer drones introduces challenges when processing imagery at scale. At HOTOSM we have been refining workflows alongside OpenDroneMap to process areas of <100 km² captured using drones such as the DJI Mini series, combining drone-specific correction techniques with scalable Kubernetes-native processing pipelines.