RETRIEVAL-AUGMENTED GENERATION AND KNOWLEDGE - AUGMENTED GENERATION FOR HUMAN-SUPERVISED CURRICULUM DESIGN

Source abstract

Universities design curricula from programme goals, learning outcomes, courses, credits, and prerequisites. Regulatory requirements are also checked during curriculum design. We present Curriculum-KAG, a human-supervised decision-support prototype for curriculum design. It sup- ports bachelor’s, master’s, and doctoral programmes. The proposed method combines Retrieval- Augmented Generation (RAG) and Knowledge-Augmented Generation (KAG). It uses Kazakhstan’s historical educational-programme corpus and multilingual retrieval. Contextual expert evidence is also incorporated into the process. A curriculum graph represents courses, prerequisites, and pro- gramme structures. Constrained semester planning is used to generate feasible curriculum plans. Archived pair-classification and ranking results are retained as separate experiments. Results from six programmes are also reported as a separate pilot study. An earlier engineering exercise produces 80 curriculum plans. Consultations with 28 experts provide additional contextual evidence. These results are reported with their respective evidentiary limitations. Neither study is treated as indepen- dent proof of academic quality. Course provenance is exposed for committee inspection and revision. Post-generation structural checks are also applied to proposed plans. The current evidence supports supervised institutional piloting of Curriculum-KAG.