Domain-Specific Modeling Languages (DSMLs) have been recognized as a viable solution for reducing the gap between domain abstractions and computational expression within specific domains. In several domains and contexts, DSMLs have been applied successfully to various areas (e.g., finance, combat simulation, and image manipulation) and have shown improvements to productivity and quality. However, development of a new DSML is not an easy task for either computer scientists or end-users because designing and implementing a DSML requires profound knowledge of the domain and deep experience in modeling language development. To address the challenges of DSML development, this doctoral symposium abstract outlines a new approach for building DSMLs that represents a demonstration-based technique for specifying the details of a new modeling language. The approach provides an environment for describing and generating the abstract and concrete syntax of a DSML. Initial work on describing the semantics of a new DSML is also a focus of the work. The research represents an investigation into a technique that allows end-users to sketch (or demonstrate) a domain model with free-form shapes. The goal of the proposed research is to develop the underlying science and tool support to enable end-users to assist in designing a DSML for their domain, while minimizing the typical mundane tasks of DSML development involving many accidental complexities.