This study , led by Dhruv Balwada, introduces a new data-driven parameterization to better represent how mesoscale eddies remove potential energy from the ocean in climate models. Unlike the widely used Gent-McWilliams (GM) scheme, which can hinder resolved eddies and lacks a robust basis for tuning, this approach is both flow-aware and scale-aware, minimizing negative impacts on resolved dynamics. Built with a lightweight neural network, the method is efficient, easy to implement, and successfully tested in NOAA’s MOM6 model. The results highlight a promising path to reduce structural errors and improve the realism of climate simulations.