Graph Neural Networks (GNNs) have become a go-to approach for learning from graph-structured data, with impressive results in areas like social networks, molecular chemistry, recommendation systems, and knowledge graphs. In this survey, we take a deep dive into the field—exploring the core concepts, different model architectures, training techniques, and real-world applications. We trace the journey of GNNs from early spectral methods to today’s message-passing models, and highlight key challenges such as over-smoothing and scaling to large graphs. We also provide a side-by-side comparison of model performance on common tasks. While GNNs have made major strides, there are still open questions—especially around working with heterogeneous and dynamic graphs, and understanding how well these models generalize in theory.