How do I learn to build and deploy generative AI models in production?

Summary

• Start with free, self-paced training, then build hands-on. The Databricks training catalog covers generative AI fundamentals and generative AI engineering, backed by ebooks like the Big Book of Generative AI.
• Build on Mosaic AI. Use Databricks AI Search for retrieval, MLflow to track and register models in Unity Catalog, and the Mosaic AI Agent Framework to build retrieval-augmented and compound AI applications.
• Deploy with Mosaic AI Model Serving. Serve custom models, Databricks-hosted foundation models, external models, and agents behind real-time and batch inference endpoints.
• Operate with MLflow and MLOps. MLflow experiment tracking, the Model Registry in Unity Catalog, and MLflow Tracing provide versioning, evaluation, and end-to-end observability in production.
• Govern centrally. Unity Catalog applies consistent access control and lineage across data, models, and agents.

How do I learn to build and deploy generative AI models in production?

Learning to build and deploy generative AI in production means following a path from foundational concepts to hands-on development, then to serving and operating models with governance and monitoring. Databricks provides that path through structured training, hands-on tutorials, and the Mosaic AI capabilities of the Databricks Data Intelligence Platform, so you can take a generative AI application from prototype to production in one environment.

Why build and deploy generative AI in production on Databricks

• A structured learning path. Begin with self-paced courses in the Databricks training catalog, including generative AI fundamentals and generative AI engineering. Reinforce concepts with the Big Book of Generative AI and, for production operations, The Big Book of MLOps.
• Build with Mosaic AI. Databricks AI Search provides a vector database for semantic search and retrieval-augmented generation (RAG). MLflow tracks experiments and registers models in Unity Catalog, and the Mosaic AI Agent Framework helps you build RAG and compound AI systems.
• Deploy with Model Serving. Mosaic AI Model Serving deploys custom models (Python models in MLflow format such as scikit-learn, XGBoost, PyTorch, and Hugging Face), Databricks-hosted foundation models, external models, and agents behind real-time and batch inference endpoints.
• Operate for production. MLflow experiment tracking and the Model Registry in Unity Catalog give you versioning and lineage, and MLflow Tracing logs each step of a generative AI application for debugging and monitoring. See MLOps workflows on Databricks for production deployment patterns.
• Prototype quickly. The AI Playground lets you test foundation models before you commit to an architecture.
• Govern centrally. Unity Catalog applies consistent access control, lineage, and governance across data, models, and agents.

Getting started

• Take the generative AI courses in the Databricks training catalog.
• Read the Big Book of Generative AI for build patterns and The Big Book of MLOps for production operations.
• Add retrieval to your app with Databricks AI Search and track models with MLflow.
• Deploy with Mosaic AI Model Serving and follow MLOps workflows.
• See the latest capabilities in Mosaic AI announcements from Data + AI Summit.

FAQs

Where should I start if I am new to generative AI on Databricks?

Start with the self-paced generative AI courses in the Databricks training catalog, then work through hands-on tutorials and the Big Book of Generative AI.

How do I deploy a generative AI model to production?

Register the model with MLflow in Unity Catalog and deploy it with Mosaic AI Model Serving, which supports custom, foundation, and external models as well as agents across real-time and batch inference.

What supports generative AI applications in production?

MLflow experiment tracking, the Model Registry in Unity Catalog, MLflow Tracing, and Databricks AI Search support building, evaluating, and operating generative AI applications, with Unity Catalog providing governance.