Build a self-scaling, event-driven OCR pipeline on Kubernetes using Small Language Models, Qwen 3.5, and the GLM-OCR SDK.