Linkedcare (领健)

Solution

Linkedcare moved its orthodontic AI inference stack from fixed GPU servers to Alibaba Cloud Function Compute so clinic demand could determine capacity instead of long-lived infrastructure guesses. The team packages the orthodontic model as a standard container image in ACR, and Function Compute pulls that image, attaches GPU resources, and starts the inference runtime only when requests arrive. Pre-reserved GPU instances provide a low-latency floor for real-time use, while elastic scale-out handles demand spikes without keeping large monthly GPU fleets online. The platform also contributes built-in monitoring, logs, and alerts, which matters because the model is under ongoing iteration and failures need to be caught before clinics report them manually. In practice, a front-end dental request triggers GPU-backed inference, returns a result for review, and then lets resources scale back after demand drops.

Data flow

Clinic-facing applications submit orthodontic requests, Function Compute pulls the algorithm image from ACR, attaches GPU resources, runs the inference workload, and returns the diagnostic output. Metrics, logs, and alerts from the same runtime give the team visibility into model behavior and scaling events.