Fluid simulations are an essential tool used in the aerospace, automotive and marine engineering sectors to analyse and test designs. However, the current state-of-the-art simulation software is slow and computationally expensive, making it a critical blocker to engineering progress. Current computational fluid dynamics (CFD) solutions force engineers into painful trade-offs between the accuracy and speed of their simulations, meaning they cannot fully explore the space of possible high-performing designs. Our innovation is to develop a new CFD product that will use machine learning to massively increase the computational efficiency of these fluid simulations. We are building upon cutting-edge machine-learning research that demonstrates order-of-magnitude efficiency gains can be achieved by developing a "fully-differentiable" solver. A differentiable solver means we can compute gradients directly through the solver to train ML models, which is not a feature in any existing commercial fluid simulation product. Currently, we have built a 2D differentiable solver that we have benchmarked against existing academic and industry-trusted CFD solutions. The results of this project will be to create a web application where engineers can configure, run and analyse the performance of designs using machine learning accelerated simulations.