This paper describes SKIMA, a mediation system that gives transparent access to heterogeneous and distributed sources considering their semantics and the semantics of application requirements. It is based on a pivot model that abstracts concepts and semantic relations based on the SHIQ(D) description logic [10]. We use this model to represent application domains and source contents. Our approach provides an integrated and global view over local sources and couples it to the description of an application domain using semantic correspondences. In order to do so, it applies on inference to reason about these correspondences and other metadata, and exploits this knowledge to perform intelligent query processing tasks. We apply on a first prototype of the ADEMS framework to validate our approach in the Computer Assisted Instruction context by configuring SKIMA for a programmed instruction system.