Predictive process monitoring (PPM) uses event logs to forecast process outcomes and behaviors. While existing approaches effectively model simple control-flow patterns, they often overlook organizational and resource information embedded in the event log. Therefore, using a design science approach, we develop TGN-AST, a novel hybrid model leveraging this context information to improve predictions. TGN-AST combines the benefits of temporal graph neural networks and transformer-based sequence modeling to predict processes with resource and organization structures. It extracts organizational relations from event logs as continuous-time dynamic graphs, learns temporal node embeddings that capture the evolution of relationships, and incorporates these in a multi-task transformer model. The evaluation of TGN-AST across multiple real-world IT Service Management event logs reveals statistically significant and incremental improvements, particularly in next activity predictions. The findings show that integrating a temporal graph network with a transformer enhances predictive performance, bridging the gap between sequential and graph-based process analysis. Furthermore, our research provides empirical evidence underscoring the utility of temporal context modeling in improving the prediction accuracy.