Despite their name, social learning networks often lack explicit support for collaborative learning, even though collaborative learning offers benefits over individual learning. The outcomes of collaborative, project-based learning can be optimized when team formation experts assemble the project teams. This paper addresses the question of how to provide team formation services to individual, self-directed learners in a social learning network so they can make use of and profit from project-based learning opportunities. A model of a team formation process is presented, based on current team formation theory. It is used to design an automated team formation service that can be used by self-directed learners to form teams for project-based learning. Starting from a project description situated in a knowledge domain, the model defines three categories of variables that govern the team formation process: (I) knowledge, (II) personality and (III) preferences. Learner data on these categories are combined in a measure of fit, which calculates the best team for a project. A novelty introduced is that, depending on the desired project outcomes the relative weight of the categories can be altered to optimise the project formation process. The feasibility of the approach is demonstrated in an example in which the proposed algorithm is used to determine the most productive team for a project. Finally, future work and research are indicated.