Databricks Mosaic AI

Databricks Mosaic AI is an integrated AI/ML platform that is part of the Databricks Data Intelligence Platform, built on a data lakehouse architecture. The platform covers the full model lifecycle: data preparation, training, fine-tuning, deployment, and monitoring. Mosaic AI was created following Databricks acquisition of MosaicML in June 2023 for approximately $1.3 billion, incorporating its LLM training and serving technology.

Key Components

The platform consists of: Agent Bricks for building AI agents optimized on enterprise data; Unity AI Gateway as a central governance hub for LLMs and MCPs; Vector Search, a native vector database with real-time sync; Agent Framework and Evaluation for building and assessing production agents; Model Serving for deploying GenAI and classical ML models; Mosaic AI Training for fine-tuning and pre-training LLMs; Managed MLflow as the MLOps backbone; and Databricks Notebooks for collaborative data science.

Lakehouse Integration

The key advantage of Mosaic AI is its deep integration with the data layer: Delta Lake, Unity Catalog (data governance), Databricks SQL, and Apache Spark. Models have native access to structured and unstructured data stored in the same ecosystem, eliminating data movement and simplifying governance. The platform runs on AWS, Azure, and Google Cloud.