Accurate estimation of building height plays an important role in urban planning and monitoring. It is also one of the key parameters in quantifying energy consumption, greenhouse gas emissions, population, urban heat island effect as well as planning transportation and telecommunication routes. Despite several existing studies, conducting large-scale building height estimation at a fine spatial resolution is a challenging task. Most of the large-scale solutions provide building heights at low spatial resolution (500 m - 90 m), raising questions about their usefulness. We propose an advanced deep learning model, T-SwinUNet, specifically designed for large-scale building height estimation at a fine spatial resolution of 10 m. The model harnesses salient features from the spatial, spectral, and temporal dimensions of the Sentinel-1 SAR and Sentinel-2 MSI time-series data. Currently, the model is equipped to generate 3D building model (LOD 1) across Europe. We are targeting to expand it to the global scale as the required input Sentinel-1 and Sentinel-2 data is available globally. In this project, we will be enhancing our model with semi-supervised learning and domain adaptation. This will involve training models on several terabytes of data. With our project, we will be able to provide a LOD 1 building model across globe. Moreover, we will be able to update it frequently as Sentinel-1/ 2 provides free global data with an update cycle of 5-6 days. Therefore this solution is not only helpful in the long run but also in short situations such as damage assessment due to natural disasters.