Early-stage aircraft structural trade studies often rely on empirical mass models and legacy trends, which can be unreliable for composite structures and non-conventional configurations. This work presents an automated parametric framework accelerating physics-based model generation and enabling data-efficient structural optimisation during conceptual design. Python modules automatically generate detailed global finite-element models (GFEMs) of composite wingboxes and representative fuselage segments, including outer geometry and structural layout parameters, such as ribs, stringers, spars, frames and stiffeners. Meshing, material/property assignment (isotropic or laminated composites) and boundary conditions and loads are automatically assigned or imported in Altair HyperMesh, enabling rapid exploration of the design space. A study on the sizing of a benchmark wingbox is performed where three layout variables (rib number, stringer number on each skin and spar-cap width) are varied throughout a Design of Experiments (DoE). The minimum-mass design is obtained for each configuration, via a composite sizing optimisation in OptiStruct featuring 75 design variables and strength and stability constraints. A multi-run strategy is also employed to mitigate convergence to local minimum solutions. As repeated optimisation runs are computationally expensive, a Gaussian Process (GP) surrogate trained on the DoE results is used to predict the optimized structural mass for each configuration. An efficient optimal configuration search is then performed using a Bayesian Optimisation loop to guide the selection of new candidate designs. Preliminary results for the combined wingbox structural layout and sizing optimisation show the surrogate capability of predicting the wingbox mass with a relative RMSE of about 0.8% and identifying near-global solutions (within 1.46%) with a fraction of computational time compared to a direct global search algorithm.