Construction Grammar (CxG) is a well-established linguistic theory that takes the notion of a construction as the basic unit of language. Yet, because the potential of this theory for language teaching or SLA has largely remained ignored, this paper demonstrates the benefits of adopting the CxG approach for modelling a student's linguistic knowledge and skills in a language tutoring application. I propose a tutoring architecture for (adult) second language learning that relies on a student model that tracks a student's constructional knowledge. This model is embodied in a fully operational student agent, which has a construction inventory, a grammar engine (to process constructions) and learning strategies (to update constructions after learning). Through linguistic interactions between a language learner and the tutoring system, the student agent is enabled to model the behavior of the real student and tries to predict his input. The student construction inventory is aligned to the real student's input after every interaction. This innovative architecture, implemented in Fluid Construction Grammar, is demonstrated here for the use case of Spanish past tense expressions, which remains a complex task even for the most advanced learners of Spanish. [For full proceedings, see ED565044.]