Shandong Xinminghui, a labor-protection equipment e-commerce company based in Linyi, has handed the matching of more than 400,000 product variants and over 300,000 upstream and downstream customers to a large language model it built itself, Qilu Network's Shandian News reported on November 13. Since AI-powered product selection went live, mis-shipments of complex categories such as respirators — once a recurring error — have fallen to nearly zero, platform transaction volume rose 10 percent year on year in the first ten months, and full-year sales are expected to top 5 billion yuan.
The transition was bumpy. The company stocks more than 400,000 types of safety equipment; respirators alone come in over a hundred functional variants rated for gases such as ammonia and hydrogen sulfide. More than 200 customer-service staff once relied on manually compiled product taxonomies and still made frequent errors. In 2023 the company spent over 20 million yuan on smart warehousing equipment, but the bundled generic software clashed with its operations — the system collapsed and orders piled up — goods worth nearly 200 million yuan sat unshipped across six months, and staff needed two months of manual re-inventorying to straighten the system out.
Policy and technology moved together. Shandong province issued "one industry, one model" guidance with subsidies in the form of computing, model and corpus vouchers; Xinminghui collected more than 6 million yuan in support over two years. Its labor-protection LLM now runs seven core modules including smart selection, demand forecasting, intelligent dispatch and supply-chain traceability, cutting costs by more than 5 million yuan a year. Shipping efficiency is up 45 percent, management costs are down by more than half, and order and account accuracy both exceed 99.9 percent.
County-level industrial belts are becoming the testing ground for vertical models. Digitization coverage among scale industrial enterprises in Linyi's Lanshan district has reached 79 percent, with full coverage targeted for 2028. Whether Xinminghui's ambition to turn its model into an industry-wide "public brain" for safety equipment materializes will depend on how deeply the system absorbs niche domain knowledge.