© 2015 Elsevier Ltd. All rights reserved. Speech and Language Pathologists have to treat a wide spectrum of disorders, working with a large set of exercises and activities to design personalized therapy plans for their patients. This paper presents an expert system designed to provide support in that labor by automatically generating therapy plans containing semiannual activities in the areas of hearing, oral structure and function, linguistic formulation, expressive language + articulation, and receptive language. The system relies on an implementation of the Partition Around Medoids (PAM) algorithm to generate clusters of subject profiles with two levels of granularity, first considering broad diagnosis terms and medical conditions, and then looking at the specific communication skills affected. The proposal has been tested in collaboration with expert pathologists from three special education institutions of Ecuador, who were about 90% satisfied with the quality of the therapy plans provided. It was found that the two-level clustering is a crucial feature to tell apart individuals who have similar speech-language limitations, but arising from different medical conditions and, therefore, requiring different treatment.