Airborne time-domain electromagnetic surveys produce large datasets that may contain thousands of line kilometers of data. The inversion of such large datasets becomes a computationally expensive process due to the repeated calculations of the forward model data and the partial derivatives for solving the least-squares inverse problem. To improve the computational efficiency of the inversion process, we use neural networks to compute the forward model data and the partial derivatives for a broad range of resistivity structures and flight altitudes for an airborne setup. Experiments show that the integration of neural network based forward data modelling and partial derivative calculations within the inversion framework opens the possibility of faster inversions with little to no loss in inversion precision.