Modern businesses have intricate data networks that connect information such as customer behavior to marketing campaigns or fraud detection. But in order to make useful AI predictions on the data, the web of data connections often has to be unraveled. A new Stanford-born startup says it has a solution that uses a new class of artificial intelligence to solve this problem.
Kumo on Thursday announced itself to the world with $18.5 million in Series A funding that it hopes will help make the software of choice for AI prediction in the “modern data stack.” ‘, a set of cloud computing tools for storing and using large amounts of files. Sequoia Capital led the round with a valuation of $100 million; other participants came from Ron Conways SV Angel and his son Ronny Conways A Capital.
The Mountain View, California-based startup was launched four months ago by founders Vanja Josifovski (formerly chief technology officer at Pinterest and Airbnb’s Homes business), Hema Raghavan (a former LinkedIn technical director), and Stanford professor Jure Leskovec , who also previously worked there, founded Pinterest’s chief scientist. The company is the result of five years of academic research conducted by a Stanford team with Leskovec in collaboration with the German University of Dortmund. They focused on an emerging form of AI dubbed Graph Neural Networks, which approaches machine learning by treating data as if it were a complex graph network. Older forms of neural networks have become good for “structured data” tasks such as image recognition or speech recognition, but are hampered by data with unordered connections.
The research led to the development of PyG, an open-source tool for learning graph neural networks that was first introduced five years ago. Meanwhile, the founders of Kumo implemented the technology on Pinterest and LinkedIn. “LinkedIn is like a big graph,” as Josifovski, the CEO, puts it, before claiming that neural graph networks “have the potential to revolutionize machine learning in much the same way deep learning revolutionized language.”
But while big tech companies have the resources and manpower to build these tools with in-house teams, most companies can’t do the same. This is where Kumo comes into play. The company’s software leverages PyG’s technology as the foundation of its software, which helps customers more easily create complex predictive models from their business data. “Today you can find out how many customers have churned after 30 days,” says Josifovski. “Kumo aims to provide the same functionality for the future – for the next 30 days.” Kumo’s product is primarily designed for data analysts and data scientists, and according to Josifovski, it should also be usable for non-technical employees. “Every company has trouble hiring data scientists,” he says. “If we’re able to package in a consumer-centric way, it’s going to have a profound impact on the computing world.”
Kumo will use the funds raised to expand product capabilities and continue to focus on research and development. The startup currently employs more than 20 people, most of them engineers from the Stanford-Dortmund network with expertise in graph neural networks. But so far, the startup hasn’t generated any significant revenue. The subscription-based product is in beta testing and will be used by “select customers,” Josifovski says, though he won’t name names and has no timeline for when the product will be available in stores. According to Konstantine Buhler, the Sequoia partner who led the financing, Kumo sought clients among the largest public market companies. “There’s a sucking sound here,” he says. “The market wants that.”
Still, Kumo will have a tall order to bring neural graph networks into the mainstream. Multi-billion dollar companies like Databricks, DataRobot and Dataiku have already built lucrative businesses with different approaches to data science. According to Josifovski, Kumo solves similar problems as some of these companies. “But we intend to make machine learning an order of magnitude easier,” he says. “We’re basically trying to skip the current state of AI and make current methods obsolete.”