In this study, we address the challenge of computationally expensive simulations of complex geometries, which are crucial for modern engineering design processes. While neural network-based flow field predictions have been suggested, prior studies generally exclude complex geometries. Our objective is to enhance flow predictions around complex geometries, which may often be deconstructed into multiple single, simple bodies, by leveraging existing data on these simple geometry flow fields. Using a case study of tandem airfoils, we introduce a method employing the directional integrated distance representation for multiple objects, a residual pre-training scheme based on the freesteam condition as a physical prior, and a residual training scheme utilising smooth combinations of single airfoil flow fields, also capitalising on the freesteam condition. To optimise memory usage during training in large domains and improve prediction performance, we decompose simulation domains into smaller sub-domains, each processed by a different network. Extensive experiments on three new tandem airfoil datasets, comprising over 2000 fluid simulations, demonstrate that our proposed method and techniques effectively enhance tandem airfoil prediction accuracy by up to 63\%.