Kumo’s Hema Raghavan: Turning Graph AI into ROI

Key takeaways

Kumo AI utilizes advanced graph neural networks to automate machine learning processes, making AI more accessible to businesses.
The power of GNNs lies in their ability to model complex relationships within data, significantly improving predictive analytics.
Kumo’s architecture works directly with relational data warehouses, enabling rapid value generation without requiring extensive data science resources.
The future of AI is centered around enhancing productivity and implementing cutting-edge technology to drive business outcomes.
Company culture at Kumo fosters innovation and employee engagement, key factors in retaining top-tier talent in the competitive AI space.

Summary

Introduction to Kumo AI and Graph Neural Networks

In this episode, Hema Raghavan, the co-founder of Kumo AI, discusses how her company is changing the landscape of AI with its innovative approach to graph neural networks (GNNs). With a solid background in AI development, particularly from her time at LinkedIn, Hema explores how GNNs work optimally with relational data stored in platforms like Snowflake and Databricks. By automating feature engineering processes, Kumo aims to provide fast, efficient predictive analytics while reducing the reliance on deep expertise in AI. Hema emphasizes the importance of GNNs over traditional machine learning methods, particularly in their ability to handle complex data relationships as graphs.

Demystifying AutoML and GNNs

The podcast touches on the evolution of AutoML, highlighting shifts in AI methodologies. Hema explains that previous AutoML frameworks relied heavily on CPU-based models and manual feature engineering, which required considerable data science expertise. Kumo's groundbreaking approach leverages GPUs alongside GNNs, allowing for automated machine learning that eliminates much of the manual work typically associated with the feature engineering process. Hema likens Kumo's functionality to that of self-driving cars, giving users the choice to either let the system operate automatically or to dive deeper into the inner workings.

Practical Applications of Kumo in Industries

Hema shares how Kumo's technology can be applied across various industries, including fintech, healthcare, and content recommendation systems. For example, Kumo can help in predicting user churn, demand forecasting, and identifying fraudulent activities. Hema provides a relatable example of how a video-streaming platform could enhance user recommendations by leveraging data on past user behavior analyzed through GNNs. This practical perspective showcases Kumo's potential to drive revenue and reduce costs for organizations by transforming data into actionable insights.

The Future of AI and Kumo’s Vision

Looking ahead, Hema is optimistic about the impact of AI on productivity and how Kumo can lead the charge in simplifying the integration of advanced AI solutions into businesses. She discusses the importance of innovations at the intersection of hardware and software, specifically through the optimization of GNNs. Hema expresses her excitement over the potential for new applications built on Kumo's framework, particularly how these could revolutionize various sectors by enabling more personalized and predictive analytics on an unprecedented scale.

Building a Robust Company Culture at Kumo

Finally, Hema emphasizes the significance of company culture, crediting the retention of top talent at Kumo to a supportive environment where individuals are empowered to innovate and collaborate. With a clear company vision focused on solving real-world problems and enhancing the role of AI, Kumo fosters an atmosphere where team members can thrive and contribute to the broader goals of the organization.

Show notes

Hema Raghavan is co-founder of Kumo, a company that makes graph neural networks accessible to enterprises by connecting to their relational data stored in Snowflake and Databricks. Hema talks about how running GNNs on GPUs has led to breakthroughs in performance as well as the query language Kumo developed to help companies predict future data points. Although approachable for non-technical users, the product provides full control for data scientists who use Kumo to automate time-consuming feature engineering pipelines.

Mentioned in this episode:

Graph Neural Networks: Learning mechanism for data in graph format, the basis of the Kumo product

Graph RAG: Popular extension of retrieval-augmented generation using GNNs

LiGNN: Graph Neural Networks at LinkedIn paper

KDD: Knowledge Discovery and Data Mining Conference

Hosted by: Konstantine Buhler and Sonya Huang, Sequoia Capital