Kumo: AI models for relational data

Your most useful business data lives in your data warehouse or relational database. Yet models struggle with relational structures, requiring time-consuming feature engineering or lossy workarounds to use the data for AI applications.

The Kumo platform lets you train highly accurate AI models for predictions and embeddings directly on relational data, with...

No feature engineering.
No feature experimentation.
No ML pipelines.

Thanks to breakthroughs in Relational Deep Learning and Graph Transformers.

Now anyone can build accurate and impactful Predictive AI applications across more parts of the business, up to 20x faster than with traditional ML methods.

12.3 Hours

<0.5 Hours

25x
Faster

Traditional ML with
feature engineeringKumo / Relational
Deep Learning (RDL)

RelBench study: 12.3 hours for an expert data scientist to build a basic predictive model using feature engineering and LightGBM or XGBoost, compared to just 0.5 hours using Relational Deep Learning.