databricks-model-serving

Deploy MLflow models to Databricks serving endpoints by logging the model, registering it in Unity Catalog, and triggering an async job-based deployment. This workflow eliminates manual infrastructure setup and provides scalable, governed model hosting with built-in authentication.

Yes, Databricks Model Serving supports deploying GenAI agents built with LangGraph and ResponsesAgent. The deployment workflow handles agents that integrate with Unity Catalog Functions and Vector Search indexes for retrieval-augmented generation and external tool calls.

Yes, Databricks Model Serving supports custom PyFunc models containing preprocessing logic alongside classical ML frameworks like sklearn, xgboost, and pytorch. You can package custom transformations and prediction logic into a single deployable endpoint.

Monitor Databricks endpoint status after model deployment using MCP tools, the Databricks SDK, or the REST API. These interfaces enable you to check deployment progress, query endpoints for real-time predictions, and verify endpoint health programmatically.

Query a deployed ML model on a Databricks endpoint for real-time predictions by sending requests through the Databricks SDK or REST API. The endpoint handles authentication and routes requests to your registered MLflow model or AI agent for immediate inference.

Yes, registering MLflow models in Unity Catalog is part of the standardized deployment workflow on Databricks. Unity Catalog provides the governance and authentication layer required before models or AI agents can be deployed to scalable serving endpoints.