This project studies how to build foundation models for graph-structured data. We explore graph-of-graphs representations on Riemannian manifolds to capture complex geometric and hierarchical structures, aiming to develop generalizable graph foundation models that can adapt across diverse graph learning tasks.