AI models that connect the dots.

Get predictions and embeddings from your relational data, without feature engineering.

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 engineering
Kumo / 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.

Reddit

Record-setting lift of user engagement models using Kumo embeddings.

databricks

5.4x lift in conversions from new lead-scoring models

Sainsbury's

Powering recommendations for UK's most well-established supermarket.

Doordash
Learn how

1.8% engagement lift across 30M+ users, fueling MAU growth with Kumo models.

Latest Research

More research

Built by pioneers of AI and engineering from Airbnb, Stanford, Google, LinkedIn, and Pinterest

Vanja Josifovski

Vanja Josifovski

CEO and Co-Founder
Former CTO at Airbnb and Pinterest

Jure Leskovec

Jure Leskovec

Co-Founder & Chief Scientist
Professor at Stanford
Co-creator of RDL and GNN
Former Chief Scientist at Pinterest

Hema Raghavan

Hema Raghavan

Co-Founder & Head of Engineering
Former Sr. Director of Engineering at LinkedIn

About the
company

Start shipping great AI applications in less time.