04_model_serving_preparation.md¶

Model Serving Preparation¶

The goal of this step is to package the registered model and all its dependencies into a deployable artifact.

1. Artifact Collection¶

The prepare_model_serving component downloads the specific model version from the MLflow Model Registry.

2. Creating the Serving Application¶

The component dynamically generates a serve.py file using FastAPI. This application includes: - /predict: A POST endpoint that accepts bike features and returns the predicted demand. - /health: A GET endpoint for Kubernetes liveness/readiness probes.

3. Dependency Packaging¶

A requirements.txt file is created containing: - mlflow - fastapi - uvicorn - pydantic - pandas

4. Final Artifact¶

All components (model/ directory, serve.py, requirements.txt, and model_metadata.json) are bundled into a ZIP archive. This archive can be used by a CI/CD pipeline or a Kubernetes deployment to spin up a prediction service.