The document discusses improving current e-learning systems by automating course composition based on semantic knowledge. It proposes creating user profiles and selecting personalized learning objects (LOs) that account for dependencies between LOs to dynamically compose courses tailored for each individual user. This would improve upon pre-defined, generic courses by cross-referencing information between courses. The document also outlines basic workflow patterns like sequence, parallel split, and synchronization that could be used for composing courses, and identifies defining goals and improving composition algorithms as areas for future work.