Development of an LLM-Based Multi-Agent System for Sewing Sequence Planning

Sewing sequence planning determines the order for assembling a garment yet is still performed manually. This study proposes an Large Language Model (LLM)-based multi-agent system to generate the sequence, addressing a task whose knowledge divides into what can be formalized and what remains with experts. General construction principles were formalized into standard production knowledge, and a semantic metadata schema was developed to specify the functional role of panels, edges, and their correspondences. Generation, verification, question, and revision agents plan the sequence within a human-in-the-loop cycle that incorporates expert feedback when a decision cannot be resolved from these formalized resources. Across 100 patterns, the multi-agent configuration improved on the single-agent configuration while expert intervention was required for some patterns. The system turns the sewing sequence from a given input into a computational outcome by resolving decisions from formalized knowledge and eliciting expert feedback where design-specific judgment is required.