Large-scale reconstructions, nanostructures, defects and moire patterns typically entail length scales of several nanometers. Their ab-initio modeling thus requires unit cells with a prohibitively large number of atoms. For example, the unit cell of magic-angle bilayer graphene, featuring superconducting many-body states, includes around 12 000 carbon atoms. Conversely, the resulting modulations in local electronic structure, strain or charge states critically affect surface properties. We combine machine learning [1] and numerical optimization techniques with small-scale DFT calculations to derive large-scale tight-binding parametrizations. I will review several approaches we used to describe the density of states, strain patterns [2], quantum transport [3], phonons and excitons [4] in moire superstructures and defects. Our toolset is general and can be applied to a wide range of other materials, enabling accurate large-scale simulations of material properties in the presence of large-scale reconstructions. [1] Machine learning sparse tight-binding parameters for defects, npj Computational Materials 8, 116 (2022) [2] Quantifying Strain in Moiré Superlattice, Nano Letters 23, 11510 (2023) [3] Stability of destructive quantum interference antiresonances in electron transport through graphene nanostructures, Carbon 214, 118358 (2023) [4] Strain control of hybridization between dark and localized excitons in a 2D semiconductor, Nature Communications 13, 7691 (2022)