The total mass of protoplanetary disks is the critical ingredient for planet formation models. Since the main gas component, molecular hydrogen, cannot be directly observed, CO lines are commonly used to estimate the total disk gas mass. In particular, less abundant CO isotopologues are more optically thin and trace the bulk gas mass. However, it is necessary to model the disk chemical evolution to infer the abundance of molecular hydrogen from CO data, making inverse modeling of observations a challenging task. We present here a novel approach to modeling line emission data based on machine learning-accelerated disk chemistry, allowing us to predict the disk chemistry without running a chemical network. To this purpose, we developed DiskCheF, a framework for disk physical and chemical modeling, and interferometric data fitting. We test this approach on rare CO isotopologue data of 7 Class-II disks, observed as part of the PRODIGE large program on the NOEMA interferometer. The machine learning-accelerated chemistry allows us to fit theoretical disk physical models to line emission data, constraining the disk gas mass and temperature distribution.