Organic energy materials are central to several emerging sustainable technologies, including organic photovoltaics, redox-active organic batteries and photocatalytic systems. Their performance is governed by a complex interplay between molecular structure, intermolecular packing, morphology, electronic excited states and environmental effects. Predictive modelling of these materials therefore requires multiscale simulations that combine realistic atomistic structures with accurate electronic-structure calculations. This project will use NAISS Medium resources to develop a GPU-focused computational workflow combining molecular dynamics, machine-learning interatomic potentials and electronic structure calculations (DFT and wave-function methods), complemented by targeted CPU-based calculations. The project will focus on three representative classes of systems: donor-acceptor blends for organic photovoltaics, redox-active organic materials in condensed phases and organic photocatalysts in solution. Large-scale molecular-dynamics simulations will be performed using GROMACS and LAMMPS to generate realistic morphologies of systems containing up to approximately 250,000 atoms. These simulations will include solution-phase equilibration, selected solvent-evaporation protocols and multiple independent replicas to capture static and dynamic disorder. The molecular-dynamics workload will be carried out primarily on Dardel-GPU. A central methodological component will be the training, benchmarking and use of machine-learning interatomic potentials to improve the accuracy and transferability of the atomistic simulations. We will explore emerging universal models such as Meta’s Universal Model for Atoms (UMA), together with system-specific fine-tuning ML-potential approaches. These models will be benchmarked against DFT reference calculations and used to assess whether they can improve the description of intermolecular interactions beyond standard classical force fields. This ML-potential layer will enable more accurate sampling of large organic condensed-phase systems while retaining computational efficiency. Representative configurations extracted from classical and ML-assisted molecular-dynamics trajectories will be investigated using electronic-structure calculations (here a special MM/QM is integrated, see Ref.1). The main quantum mecahnics calculations production workflow will use Q-Chem 6.4 together with the BrianQC module on Arrhenius GPU resources (possibility of using also VeloxChem). These calculations will provide redox properties, frontier electronic levels and excited-state information for selected configurations, enabling the role of morphology and local environment to be quantified. Complementary CPU resources on Arrhenius and Dardel will support calculations that are not optimal for GPU execution, including selected Q-Chem, Gaussian and VeloxChem calculations, solution-phase optimisations, vibrational analyses, DFT reference data for ML-potential validation and higher-level excited-state calculations such as ADC(2) with Turbomole for selected non-fullerene acceptor systems. The overall objective is to establish structure-property relationships linking morphology, environment and molecular design to electronic, optical and redox properties in organic energy materials. By combining GPU-accelerated quantum chemistry, GPU-accelerated molecular dynamics, ML interatomic potentials and complementary CPU-based electronic-structure calculations, the project will deliver a focused and efficient Medium-scale workflow. The results will provide mechanistic insight and practical design principles for next-generation organic materials for sustainable energy technologies. 1. Journal of Physics: Energy 7, 045001 (2025)