The objective of this work is to propose a system that generates learn- ing scenarios for serious games keeping into account the learners’ profiles, pedagogical objectives and interaction traces. We present the architecture of this system and the scenario generation process. The proposed architecture should be, insofar as possible, independent of application domain; it means that this system would be suitable for different domains and serious games. It's why we identified and separated different types of knowledge (domain concepts, pedagogical resources and serious game resources) in a multi-layer architecture. We also present the evaluation protocol used to validate the system i.e. the method used to generate a learning scenario and the knowledge models associ- ated with the generation process. This protocol is based on comparative method that compares the scenario generated by our system with that of the expert. The results of this evaluation, conducted with a domain expert, are also presented.