ReVisor: A Reflective Design Tool for Instructional Designers to Improve Teacher Training Materials via AI Discussions

Assessing the real-world impact of instructional design requires nuanced, in-situ data, yet such data like classroom discourse defies systematic analysis due to its qualitative intricacies. We present ReVisor, a reflective tool that helps instructional designers iteratively refine training materials by (1) analyzing classroom transcripts, (2) identifying ambiguous or misaligned applications of pedagogical strategies via multi-agent LLM discussions, (3) generating concrete revision suggestions, and (4) providing simulated real-time feedback on edited materials. We evaluate ReVisor using a benchmark that treats agent disagreement as a proxy for ambiguity and a user study with instructional designers (n=10). We measure training material improvement by reduced ambiguity in AI classification, with unresolved agent disagreement indicating unclear pedagogical guidance. Results show that ReVisor supports data-grounded, logically structured revisions that better bridge theory and practice, and contributes a computational framework for integrating AI-driven reflection into the instructional design lifecycle.