Kaskada Inc., a Seattle-based artificial intelligence startup led by Google LLC and Amazon Web Services Inc. veterans, today announced that it has raised $8 million in funding.

Kaskada is on a quest to automate the so-called feature engineering phase of enterprise AI projects, which is simultaneously one of the most important and one of the most time-consuming parts of the workflow. Feature engineering involves extracting key details from raw training data to help the machine learning model being trained identify meaningful patterns.

A hypothetical software team building an AI that analyzes airline delays might use a log of departures, arrivals and average flight times to teach the model how to catch planes running late. The team would merge these three data points into a single, more concise data point, the number of minutes each flight is delayed, so that the AI can figure out what it’s supposed to look for more easily. These kinds of abstractions are known as features and help improve the accuracy of machine learning models while enhancing their hardware efficiency.

Machine learning projects usually employ a combination of data scientists and data engineers, with scientists focusing on such factors as statistics, correlation and causation variables and data engineers building software to process the data sets the scientists recommend. The two disciplines require different skills and the people often work in separate groups, said Chief Executive Davor Bonaci.

“Data scientists have a hard time figuring out factors like load times and processing speed while data engineers aren’t skilled in understanding predictive data,” he said. “Typically, data scientists toss data over the wall and the data engineers rewrite the features before they are deployed. There isn’t software to make them work well together.”

Kaskada’s namesake platform enables AI teams to centralize the entire feature engineering process in one place. Developers can pull data from their company’s internal systems of record for use in model training, cleanse the information, build features and deploy those features to production.

One way Kaskada has sought to set itself apart from the rivals is by providing deep support for real-time data. The startup’s platform can generate features based not only on historical records, like customer purchase records in a MySQL deployment, but also real-time data from stream processing systems such as Amazon Kinesis. After feature is created based on a real-time data stream, Kaskada keeps updating it with the new information that comes in.

The platform stores all of a software team’s features in a central repository accessible via an application programming interface. This arrangement allows developers to share features with one another and across teams, see how code written by a colleague works and measure components’ performance to determine which one works best for a particular project.

Kaskada will use the $8 million in capital to ramp up sales efforts around the platform ahead of its planned release into general availability later this year. To hit its goal of launching the solution before July, the startup will also hire more engineers. 

Kaskada is led by former Google Cloud engineers Bonaci and Ben Chambers, the startup’s original two co-founders, together with vice president of product Emily Kruger, who joined last year after a stint at AWS as a project manager. The startup’s investors include Voyager Capital LLC, NextGen Venture Partners LLC, Founders’ Co-op Management LLC and Walnut Street Capital Fund. 

“Kaskada is addressing a huge market need because nearly every company today is pouring significant resources into their data science efforts and very few are seeing results that meet their expectations,” said James Newell, Voyager Capital’s managing director.

With reporting from Paul Gillin

Image: Unsplash