In this paper, we describe the user modelling mechanism of AUTO-COLLEAGUE, which is an adaptive Computer Supported Collaborative Learning system. AUTO-COLLEAGUE provides personalised and adaptive environment for users to learn UML. Users are organized into working groups under the supervision of a human coacher/trainer. The system constantly traces the performance of the learners and makes inferences about user characteristics, such as the performance type and the personality. All of these attributes form the individual learner models, which are built using the stereotype theory. User modelling is applied in order to offer adaptive help to learners and adaptive advice to trainers aiming to support them mainly in forming the most effective groups of learners.