Designing a hybrid powertrain remains a complex task. It is an intricate system involving numerous variables that are spread over different levels: architecture, component technologies, sizing, and control. There is currently a lack of frameworks or tools that help in exploring the entire design space and in finding the global optimal solution throughout these levels. This article proposes a systematic methodology that tries to answer a part of this need. Starting from a set of chosen components, the methodology automatically generates all the possible graphs of architectures using constraint-programming techniques. A tailored representation is developed to picture the graphs. They are then transformed into other types of representation (tables describing the connections and the powertrain modes). Based on these representations, the architectures are automatically filtered and the most promising ones are selected. They are automatically assessed and optimized using a specifically developed general hybrid model that calculates the performance and fuel consumption of all the generated architectures. This model is inserted inside a bi-level optimization process: a Genetic Algorithm is used on the sizing and components level, while the Dynamic Programming is used on the control level. A case study is performed and the capability of the methodology is proven.