In this session we will detail the process of leveraging open geospatial data sources (specifically, OpenStreetMap and OpenAerialMap) for impactful geospatial machine learning research. We will also discuss potential opportunities as well as limitations of open geospatial data and how to overcome them. We will then walk through our submission to the 2024 I-GUIDE Spatial AI Challenge, where we built a pipeline that converts raw OSM building labels and high-resolution imagery into usable training data. Using this pipeline, we trained a model to classify building roof materials and shapes, establishing performance baselines for a problem space that remains underexplored yet has broad applications in energy planning, urban climate studies, and disaster resilience.