This thesis explores the constraints of self-assessment creation in higher education, focusing on the lack of tools, standardization, and time. We conduct didactic field expert interviews to investigate how teaching evolves towards learning management systems (LMS) with new possibilities for selfassessment generation. We propose EvalQuiz, a novel system that enables language model-assisted self-assessment creation. The system generates questions of different types, improving upon related work while addressing language model restrictions through filtering. EvalQuiz aims to bridge the gap between lecturers and students, enabling lecturers to work with educational objectives and lecture materials to provide student-centered self-assessment. The thesis aims to answer questions about the reliability and quality of self-assessment generation. EvalQuiz proposes a message composer scheme to reliably generate output according to a specification. The thesis defines a standardized self-assessment specification used with EvalQuiz. We conduct a survey evaluating EvalQuiz’s real-world performance.