Deep generative models of 3D shapes have sparked a lot of attention in the scientific community. Nonetheless, they almost all produce discrete shape representations like voxels, point clouds, and polygon meshes. We provide the first 3D generative model for a fundamentally different shape representation: expressing a shape as a series of CAD procedures. CAD models, unlike meshes and point clouds, encode the user creation process of 3D shapes, which are widely utilised in industrial and engineering design. Existing 3D generative models, on the other hand, have considerable hurdles due to the sequential and irregular structure of CAD procedures. We propose a CAD generative network built on the Transformer, based on an analogy between CAD processes and spoken language. To encourage future research on this topic, we have made this dataset freely available. The 2D foundation of parametric Computer-Aided Design (CAD), the most common modelling paradigm for manufactured items, is engineering sketches. In this study, we look at the challenge of learning-based a deep engineering sketch creation as a first step toward parametric CAD model synthesis and composition.