Question
A backend service uses the MLflow Deployments SDK to query a custom Databricks endpoint. It works from a developer notebook but fails in the production container because the Databricks host and token are not configured. Which design should the team use?