How HOT uses open source on AWS to power humanitarian AI

This article describes how the Humanitarian OpenStreetMap Team (HOT) built fAIr, an open source AI-assisted mapping tool on AWS that accelerates disaster response by automating geographic feature detection from satellite imagery.

• fAIr uses AI models to pre-generate building footprints for human review instead of manual tracing, significantly speeding up mapping efforts
• Architecture combines Amazon S3, PostgreSQL with CloudNativePG, ZenML MLOps on Amazon EKS, and Karpenter for GPU provisioning
• Inference uses Knative serverless workload management on CPU-based EC2 instances for low-latency predictions without permanent GPU costs
• Infrastructure deployed via Helm Charts and ArgoCD for reproducible, consistent environments
• Venezuela earthquake response: fAIr delivered building density and damage assessments within 24 hours; 451 volunteers digitized 53,000 buildings in one week
• Climate Ready Cities program uses fAIr to map transit, drainage, and accessibility in 13+ countries for resilience planning
• Map4Mangrove project in Indonesia monitors mangrove rehabilitation with drone imagery and geotagged field data integrated via UMap dashboard

HOT demonstrates how combining open source tools, AWS cloud infrastructure, and AI enables communities to rapidly produce high-quality geospatial data for disaster response and climate adaptation.