Quantum machine learning is rapidly advancing and has already found several applications from chemistry to finance. In this work, we propose scalable hybrid quantum graph neural networks and quantum graph attention networks designed for the NISQ-era. We apply these networks to the jet tagging task and demonstrate that the proposed models achieve competitive performance compared to classical models. Further, we also discuss certain experiments conducted on the Pythia8 Quark-Gluon jet dataset to identify key features and obtain compact representations of the dataset. We also verify that these features strongly influence the performance of the models and enable significant improvements in performance even with very small models ( \(\approx \) 200 parameters). Through this work, we also take a step forward in the interpretability of complex models like ParticleNet and Quantum Transformers in the context of jet tagging.