Many critical decisions of any design process occur in the early exploratory stages when a design is in its most rough form. This project presents a process to enhance this early stage design exploration by making it easier and more intuitive to explore a wider range of possibilities. It’s called V-Sketch. We employ digital computation to translate from intuitive sketching to 3D geometric elements. Unlike earlier work, we decouple the sketch analysis from the creation of the 3D forms allowing for a modular and therefore more flexible representation of the resulting elements. This is accomplished by introducing an intermediary, qualitative description of the sketches. "Sketch analysis" using machine learning techniques is employed to generate this intermediate description of the sketch data as the designer draws. A "reconstruction function" then translates the description into a resulting 3D form. The introduction of a reconstruction function allows an artist or designer to flexibly redefine the solution space of possible 3D forms from the same sketch as they desire without having to retrain the machine learning algorithms. Our methods are focused primarily on architectural design but potentially generalizes to other visual design problems such as, sculpture, virtual environments and industrial design. This thesis presents a conceptual framework for the V-Sketch system. We present the interactive construction of an open source data set of labelled three-dimensional drawing data, and we show a prototype implementation of the system that demonstrates the approach and motivation for the project.