Xiangnong Technology (香侬科技)

Solution

Xiangnong Technology built its Chinese NLP platform on Alibaba Cloud instead of operating its own machine room. The company first moved workloads to Elastic Compute Service and self-managed Kubernetes, then adopted AKS as the business expanded so containerized model services could scale with less operations overhead. For deep-learning training and inference, Xiangnong uses GPU and AMD instances, with Object Storage Service acting as the shared corpus and artifact layer for large training sets. EMR handles preprocessing and analytics on raw language data before jobs are scheduled into the cloud AI system, while Cloud Monitor watches GPU utilization and alerts on runtime anomalies. Load balancing, autoscaling, and orchestration complete the delivery path so the same stack can support FireDragon Writing, intelligent document processing, smart Q&A engines, public-opinion monitoring, and public-sector applications such as the Yuhang District People's Congress suggestion-routing platform.

Data flow

Enterprise text, public-opinion, and domain datasets are stored in OSS, preprocessed in EMR, and then fed into GPU and AMD training jobs running on ECS and AKS. The trained models are exposed through load-balanced application services that power writing assistance, document processing, smart Q&A, and public-sector routing workflows, while Cloud Monitor feeds alerts and capacity signals back to the operations team.