 This paper proposes a novel approach for predicting the number of positive and hospitalised cases due to Covid-19. It uses a multi-scale graph neural network to integrate population mobility data with fine scale geographic zones of a few thousand inhabitants. This allows it to capture the interactions between areas and accurately estimate the number of positive and hospitalised cases. Additionally, the model can be used to predict hospitalisation, which is important for healthcare planning. This article was authored by Konstantinos Skianis, Giannis Nicolenzos, Benoar Galix, and others.