Oxide perovskites constitutes and important class of materials with a wide range of properties such as ferroelectricity colossal magnetoresistance, electronic and/or ionic conductivity. These wide range of properties allows for applications ranging from electrolyzers to ferroelectric ceramics. To increase our current understanding of these materials and, more crucially, allow fine tuning of the properties of these materials we need to understand the dynamics of them at atomic level. Therefore, in this project we aim to develop neuro-evolution potentials (NEP). These models are trained on density functional theory (DFT) data, retaining the accuracy of the underlying first-principles method at a fraction of the computational cost. The model further allows us to access times scales far beyond what is accessible with DFT. The neuro-evolution potential could also allows us to learn tensor properties of the material that can subsequently be used for theoretical spectroscopy that could shine light on experimental disagreements. GPU acceleration is crucial to this project since it allows for rapid model development. Furthermore, it unlocks the possibility to create ensemble models that allows for better predictive power and active learning. This gives insight to the model performance and ultimately allows us to create better models for more accurate descriptions of the materials of interest.