Automotive software development is facing a fundamental transformation from a rigid embedded mechatronic paradigm to a flexible service-oriented model. Such a transformation is at the heart of delivering more value using the software-defined, digitalized, connected, automated and electrified vehicle of the future. Unlike modern web technologies which were readily able to adapt to service-orientation, automotive software technologies have been slower to transform. In this project we aim to use deep learning to train a model to understand the principles of software architecture from data such as source code and architectural descriptions in natural/structured languages. In partnership with human architects, such a domain understanding model would then be used to (1) Transform existing code that follows one set of architectural conventions to another set of conventions and (2) Assist coders in complying with a defined set of architectural conventions as they write new code. Such a model would therefore address the critical need of ensuring continuous architecture compliance, helping companies to rapidly develop new functionality while minimizing technical debt and the cost of maintenance.