Machine-learning-pontentials trained upon ab initio molecular dynamics datasets are used to carry out molecular dynamics simulations of diffusion in bulk and on surfaces, nucleation, interaction with environmental gases, phase segregation, phase transitions, mechanical behavior of defect free and defective crystals, and electronic properties of ceramics and intermetallics. All theoretical preditions are subject to experimental validation. Using NAISS resources (acknowledged), during 2023-present we published 29 peer-reviewed articles in high-quality journals including Science Advances (2023 & 2025), Nature Communications (2023), npj Computational Materials (2024), Nature Synthesis (2024). The availability of NAISS resources has been fundamental for our publications. The full list of publications and academic achievements can be found in the report.