Intelligent Procurement Platform Based on Drawing Analysis and Structured Data: 15% lower operating cost and zero misorders

In many small and midsize manufacturing businesses, procurement and administrative work depends on fragmented unstructured materials such as email, messengers, and spreadsheets. Drawing interpretation and material estimation often rely on personal experience and manual reading, which creates constant risk of dimension and tolerance mistakes that lead to misorders. Limited real-time visibility into cost and inventory also leaves off-cuts unused and turns hidden loss into a persistent operational burden.

To transform that process, the project introduced an intelligent decision-support platform that reads and connects drawing-centered unstructured data with AI. VLM was used to detect dimensions, tolerances, and symbols directly from drawings and convert them into structured data, while multimodal OCR and LLM pipelines turned receipts and invoices into accounting-ready records. On top of that, a Digital BOM, off-cut-priority allocation logic, and optimal vendor recommendation functions were orchestrated through a multi-agent system for intelligent end-to-end manufacturing support.

The platform raised dimension and tolerance recognition accuracy in drawing analysis to above 95% and saved about two hours of repetitive administrative work per day. By actively using leftover inventory and recommending cost-optimal suppliers, it reduced operating cost by more than 15% and effectively eliminated misorder cost. In doing so, it converted drawings, documents, and procurement history into strategic digital assets and lowered the barrier to AI transformation for resource-constrained manufacturers.