This paper presents the TensorFlow Graph Visualizer, a tool for understanding deep learning model architectures through interactive exploration of their dataflow graphs. The analytical workflow proceeds through progressive graph simplification, automated pattern detection, and iterative interactive exploration. Graph Simplification. A raw dataflow graph containing thousands of heterogeneous operations is progressively simplified through two stages: first, non-critical operations (constants, summaries) are extracted and encoded as embedded icons while remaining operations are grouped into a hierarchical clustered graph using namespace annotations, with edges bundled along the hierarchy. Second, high-degree non-critical nodes (variable declarations, bookkeeping hubs) are extracted using degree-based outlier detection, preserving connectivity information. Pattern Detection and Feature Computation. Repeated modular structures are automatically detected across group nodes using heuristic subgraph similarity checking, and quantitative properties (tensor sizes, device allocation, compute time, memory usage) are computed for visual encoding. Layout and Visualization. A hierarchical flow layout is computed recursively per cluster subgraph, and an interactive overview is rendered showing top-level model components with colour-coded repeated modules, an auxiliary node panel, and tensor-size edge encoding. Initial Reading. From the overview, the analyst immediately recognises high-level architectural features — top-level component organisation, colour-coded repeated module groups, and main-versus-auxiliary structure — forming an initial understanding before detailed exploration. Interactive Exploration. The analyst iteratively explores the model by expanding clusters to reveal nested structure, inspecting individual node details, switching between quantitative overlays, and recognising further architectural patterns including layer organisation, modular composition, and data flow paths. This cycle continues until the analyst has sufficient understanding of the model architecture. Knowledge Synthesis. The analyst formulates a coherent understanding of the model's structure, modular composition, and operational organisation for development, debugging, or communication.