Capstone Model Serving
Introduction
The model serving layer is how your models become useful — it takes prediction requests
and returns results. You'll deploy the FastAPI serving application to Kubernetes,
configured to load models from MLflow and scale based on demand.
Why This Matters
Training pipelines run periodically. Serving runs constantly, handling real-time requests.
The requirements are different:
- Low latency (users are waiting)
- High availability (downtime = no predictions)
- Scalability (handle traffic spikes)
- Observability (what's happening in production?)
This is where your Kubernetes knowledge pays off: Deployments for reliability,
Services for networking, HPA for autoscaling, health probes for self-healing.