In the course of a mere 14 months, Seattle startup GraphLab Inc. has gone from a computer science professor and a few colleagues creating open source software to analyze graph datasets to a 25-person company with a new, full-fledged system for building predictive applications that draw on a range of data types. Their customers include Exxon and Pandora.
GraphLab, which grew from the machine learning work of co-founder and CEO Carlos Guestrin while at Carnegie Mellon University, raised $6.75 million from Madrona Venture Group and New Enterprise Associates last May. Business and technology luminaries from around Seattle, including Jeff Bezos, joined together to recruit Guestrin to the University of Washington in 2012, some in the hope that it would mean bringing a promising startup here, too.
We caught up with Guestrin, just back from a GraphLab Conference in San Francisco that drew more than 700 people, to find out what this sought-after professor-entrepreneur and his growing team have been up to. He seemed genuinely surprised by the speed at which his company has grown.
“Every time I’ve set a growth target, we’ve exceeded it in a tremendous way,” he says.
GraphLab just released its first product, called GraphLab Create. It’s designed to do something lots software makers large and small are trying now: Make it easier for companies to get value from the proliferation data they have at their disposal.
Specifically, GraphLab Create lets engineers and data scientists build so-called predictive applications.
“These are things that take on historical data and sensor interaction data to make real-time predictions and decisions,” Guestrin says.
Predictive applications underpin things like e-commerce recommender systems, fraud detection programs, human sensing, and personalized medicine.
Guestrin says these applications are traditionally difficult to build and scale-up. They often require cobbling together a variety of tools that may not be compatible. And they require skilled, sought-after data scientists, who often end up being marginalized because their software code must be translated by other engineering teams, delaying implementation.
“We’ve been focusing on building a system that allows you to go from initial inspiration all the way to production with exactly the same code,” he says.
He draws an analogy to what Apple did for mobile apps six years ago with iOS.
“The space of phone applications was extremely fragmented a decade ago, and it was really hard to build new applications for the phone and deploy and sell them. When iOS came out, it made it very easy to be creative with the phone,” he says.
(And yes, like Apple, a marketplace for these apps is on GraphLab’s long-term roadmap.)
About 100 companies used a test version of GraphLab Create beginning in March. GraphLab now has several companies—including giants like Exxon, Pandora, and Stumble Upon—using it in production, generating early revenue for the company, Guestrin says. The industries GraphLab is targeting include retail, financial services, healthcare, oil and gas, marketing, and government.
GraphLab was originally started to build upon the open source technology Guestrin and his group at Carnegie Mellon created to help extract and analyze relationships between entities in the enormous, multi-dimensional graph datasets that underlie online services such as Facebook.
The company continues to offer the core of its technology open source, and participates and supports the active community that grew up around it. But after working closely with potential customers, Guestrin realized they need a comprehensive offering that goes beyond graph datasets.
GraphLab Create can also handle tabular, image, and text data.
GraphLab was housed for its first year at the University of Washington’s New Ventures Facility in Fluke Hall, where Guestrin had to ask that the standard 6-foot desks be replaced with 4-footers so he could squeeze in more people. The company moved to new digs in the Fremont neighborhood in June. There’s enough room for the current 25-person team, and enough work for many more.
“We could definitely double that and still be busy,” Guestrin says.
