A Syllabus-Grounded Framework for Bloom’s Taxonomy-Guided Question Paper Generation using Large Language Models

Setting up a quality question paper is a challenging task. Balancing syllabus coverage, cognitive difficulty, and assessment structure across multiple courses places a recurring burden on faculty—one that existing tools only partially address. This paper presents a syllabus-grounded framework for question paper generation using large language models (LLMs), where syllabus units, reference material, and blueprint constraints serve as explicit inputs to a structured, multi-stage pipeline. Rather than generating papers in a single prompt, the framework decomposes the process into discrete stages: syllabus analysis, Bloom’s Taxonomy-guided question bank generation, blueprint based assembly, and staged human review before finalization. The framework is examined through a small-scale case-based evaluation using real course syllabus inputs across ten university-level courses, with expert-style review and direct comparison against one-shot generation. In the evaluated setting, staged syllabus grounded generation produced more consistent syllabus coverage and better blueprint compliance than one-shot prompting, while keeping educators in control of the finalized question paper.