Frictional contact between deformable solids is an inherently multiscale phenomenon, in which macroscopic resistance to sliding emerges from complex interactions at the microscale, including surface roughness, material rheology, and history-dependent contact kinematics. Direct numerical simulation of such interfaces, while physically accurate, is computationally prohibitive when repeated evaluations are required for parametric studies, uncertainty quantification, or large-scale simulations. In this work, we present a multiscale modeling framework that couples micromechanical contact homogenization with physics-augmented neural networks to efficiently represent effective frictional behavior. At the microscale, detailed frictionless contact simulations are performed using a mortar-based finite element formulation, enabling robust treatment of non-conforming meshes, complex surface topologies, and viscoelastic material response. Although local tangential tractions vanish, an apparent macroscopic frictional resistance emerges naturally from the geometrical distribution of micro-contact normals. This response is systematically homogenized to obtain an effective friction coefficient as a function of surface morphology, normal pressure, and sliding velocity. To bridge the micro–macro scales, we introduce a thermodynamically consistent neural surrogate based on a convex–monotone dual-potential formulation. The neural architecture embeds physical admissibility directly into the model, guaranteeing non-negative dissipation while allowing complex velocity-dependent behavior, including both strengthening and weakening regimes. The resulting surrogate accurately reproduces the homogenized friction response at a fraction of the computational cost of direct simulation. The proposed framework provides a scalable pathway for incorporating microstructural contact physics into macroscopic models, enabling efficient and physically consistent simulation of frictional interfaces. Potential applications include tribological systems, soft-material interfaces, and multiscale contact problems in engineering and biomechanics.