Introduction

Databricks users can now deploy Kumo as an application running inside of their native Databricks environments. This allows you to use Kumo's advanced machine learning platform that leverages graph neural networks (GNNs) to deliver predictive analytics and insights directly from your relational data—all while retaining complete control of your data inside Databricks.

With Kumo as a native Databricks application, your organization can realize the following benefits:

Build on Databricks

Kumo’s seamless integration with the Databricks data platform allows you to reuse existing Databricks workloads like ETL and EDA and code. By directly utilizing your Unity Catalog (UC) and existing Databricks workflows, you can leverage the same fine-grained security mechanisms (e.g., masking fields, limited group access), built-in data privacy and compliance controls, and collaboration tools—all without data leaving your Databricks environment.

Improve Model Performance

Kumo improves accuracy by eliminating hand-crafted feature engineering, instead learning embeddings directly from raw relational data, while providing advanced controls that enable data scientists to add business value at each step. And when used in conjunction with the Databricks Photon query engine, you can access best-in-class compute and analytics runtime, as well as leverage Databricks Mosaic and Model Serving for unified building, deploying, and monitoring of LLMs.

Faster Time to Predictive Value

Kumo learns directly from your Databricks tables, eliminates time-consuming feature engineering, and uses AI to generate a model in hours, not months. Furthermore, Kumo’s automated pipelines keep models fresh so the AI always learns from the latest Databricks tables. And with MLflow as a fully managed service in your Databricks workspace, you get model management and orchestration alongside your Kumo GNN-powered predictions.