How do I start building generative AI applications on a data platform?

Summary

• Start in the AI Playground. Explore and compare leading foundation models with your own prompts — no code — then move to code with the Foundation Model APIs.
• Build with Mosaic AI. The Mosaic AI Agent Framework builds production RAG and agent apps with built-in evaluation, custom LLM judges, and MLflow tracing.
• Ground answers in your data. Databricks AI Search provides managed vector indexing with hybrid retrieval, syncing automatically from Unity Catalog tables.
• Operate with MLflow. MLflow manages the generative AI lifecycle, tracing every retrieval and generation step for debugging and monitoring.
• Govern from day one. The AI Gateway adds access control, rate limiting, payload logging, and guardrails, all under Unity Catalog governance.

How do I start building generative AI applications on a data platform?

The fastest way to start is on a platform where your models, data, and governance already live together. On the Databricks Data Intelligence Platform, the Mosaic AI toolset gives you a no-code entry point for prototyping, code-first APIs for building, managed retrieval for grounding answers in your data, and MLflow plus the AI Gateway for evaluating, governing, and operating what you ship.

Why Databricks for building generative AI applications

• A no-code starting point. Open the AI Playground to explore leading foundation models side by side with your own prompts before writing any code.
• Code-first model access. The Foundation Model APIs provide programmatic access to curated models so you can start building immediately.
• Production RAG and agents. The Mosaic AI Agent Framework builds retrieval-augmented generation and agent applications with parameterized chains, built-in evaluation metrics, custom LLM judges, and MLflow tracing.
• Retrieval grounded in your data. Databricks AI Search offers managed, scalable vector indexing with hybrid dense and sparse retrieval, and indexes sync automatically from Unity Catalog tables.
• Lifecycle management with MLflow. MLflow traces every retrieval call and generation step for end-to-end debugging, evaluation, and monitoring.
• Governance and safety built in. The AI Gateway adds access controls, rate limiting, payload logging, and guardrails for filtering inputs and outputs, all under Unity Catalog governance.

Getting started

• Prototype in the AI Playground, then call the Foundation Model APIs.
• Build a RAG app with the Mosaic AI Agent Framework and Databricks AI Search.
• Read Creating high-quality RAG applications with Databricks and the Big Book of Generative AI.
• Instrument with MLflow and govern with the AI Gateway.

FAQs

Where should a beginner start building generative AI apps on Databricks?

Start in the AI Playground to test foundation models with your own prompts, then move to the Foundation Model APIs and the Mosaic AI Agent Framework to build a production application.

How do I ground a generative AI app in my own data?

Use Databricks AI Search for managed vector indexing with hybrid retrieval; indexes sync automatically from Unity Catalog tables so answers can be grounded in your governed data.

How do I evaluate and monitor a generative AI application?

MLflow traces every retrieval and generation step and supports evaluation with built-in metrics and custom LLM judges, both during development and in production.