SpecGuard AI Manufacturing Order Review

Manufacturers often receive one order across buyer messages, technical PDFs and size spreadsheets. Conflicting quantities, colors or materials can be missed during manual review, creating avoidable production errors. SpecGuard AI is a working application that helps order coordinators compare these documents before production. Built with Python and Streamlit, it uses AMD Token Factory with MiMo-V2.6-Flash to extract manufacturing facts alongside source quotes and locations. Schema validation and independent checks for supported templates help identify incomplete or unsupported results. Deterministic rules then compare requirements and flag discrepancies. The demonstration uses three synthetic hoodie documents: the buyer requests 300 black hoodies, the technical PDF specifies 280 in charcoal, and the size spreadsheet totals 300. SpecGuard flags the quantity, color and size-total conflicts, preserving evidence for review. Users can record confirmed corrections and reviewer notes without overwriting original facts, then export JSON, CSV and clarification reports. READY means the configured checks have no remaining blockers; independent human production approval is still required. Validation includes 99 passing automated tests, 27 synthetic regression cases and a successful hosted AMD inference smoke test. These results demonstrate controlled behavior rather than general real-world accuracy. Next steps include manufacturing pilots, authenticated reviewers and persistent review records.