Internal tidal dissipation provides approximately half of the mechanical energy required for the diapycnal mixing to sustain the global meridional overturning circulation. A parameterization of the Internal tidal mixing is therefore critical for both ocean and climate models. The redistribution of internal-tide energy is strongly modulated by the background eddying circulation as well as the exchange of energy with subtidal meso- and submeso-scale eddies. To reflect the influence of these processes on internal tide dissipation, we develop a state-dependent hybrid parameterization of remote internal tidal mixing with machine learning support. We first build a global internal-tide energy dataset using a global simulation dataset, which resolves internal waves and submesoscale dynamics. A U-Net deep-learning model is then trained that ingests climate-scale-resolution circulation fields to predict the horizontal distribution of internal-tide energy and dissipation. These predictions form the core of a parameterization that responds to evolving circulation states and represents two-way coupling between internal tides and the subtidal circulation. In climate-model experiments, the parameterization produces a deeper and stronger AMOC, demonstrating that mesoscale modulation of internal-tide mixing can project onto large-scale circulation and climate.