Abstract We propose a method for reducing the spatial discretization error of coarse computational fluid dynamics (CFD) problems by enhancing the quality of low-resolution simulations using a deep learning model fed with high-quality data. We substitute the default differencing scheme for the convection term by a feed-forward neural network that interpolates velocities from cell centers to face values to produce velocities that approximate the fine-mesh data well. The deep learning framework incorporates the open-source CFD code OpenFOAM, resulting in an end-to-end differentiable model. We automatically differentiate the CFD physics using a discrete adjoint code version. We present a fast communication method between TensorFlow (Python) and OpenFOAM (c++) that accelerates the training process. We applied the model to the flow past a square cylinder problem, reducing the error to about 50% for simulations outside the training distribution compared to the traditional solver in the x- and y-velocity components using an 8x coarser mesh. The training is affordable in terms of time and data samples since the architecture exploits the local features of the physics while generating stable predictions for mid-term simulations.