Databricks Artificial Intelligence

Databricks Artificial Intelligence is the AI product area of the Databricks Data + AI Platform, made by Databricks Inc. It lets teams securely connect enterprise data with any AI model to build agent systems, and covers data preparation, agent building, deployment, evaluation and governance.

What It Is

The page describes an AI agent system as more than a single model: it combines LLMs, classical ML models, enterprise data and tools, with built-in evaluation and governance. Databricks positions the offering for enterprises that want domain-specific agents grounded in their own data rather than generic AI outputs.

Tools in the Agent Lifecycle

• Agent Bricks builds agents grounded in enterprise data, with synthetic data, custom evaluation and automated tuning.
• Document Intelligence parses, extracts and classifies enterprise documents.
• AI Search is a vector database with real-time syncing of source data, used to create vector indexes for RAG.
• AI Gateway manages models such as OpenAI GPT-4, Meta Llama 3 and Anthropic Claude 3, and supports authoring and sharing AI tools.
• Unity Gateway is described as a control plane for access, security and cost across models, agents, MCPs and tools.
• Model Serving deploys agents, GenAI and classical ML models.
• Model Training fine-tunes open source LLMs, pretrains custom LLMs or builds classical ML models.
• Managed MLflow supports MLOps and LLMOps.
• Data Quality Monitoring tracks anomalies and freshness.

Evaluation and Governance

Built-in agent evaluation uses AI judges and human feedback to measure output quality, trace root causes of issues, test fixes and redeploy. Unity Catalog governance applies guardrails, access controls and rate limits, and tracks lineage from data to deployment, including for models hosted outside Databricks.

Getting Started

Databricks offers a free self-paced GenAI training course with a completion certificate and a get-started tutorial that takes an agent from proof of concept to production. Databricks also publishes customer stories with its own reported results, which are vendor claims.