Graph convolutional networks (GCNs) are widely used for semi-supervised node classification on graphs today. The graph structure is however only accounted for by considering the similarity of activations between adjacent nodes, in turn degrading the results. In this work, we augment GCN models by incorporating richer notions of regularity by leveraging cascades of band-pass filters, known as geometric scatterings. We introduce a new hybrid architecture for the task and demonstrate its potential on multiple graph datasets, where it outperforms leading GCN models.