As companies increasingly rely on machine learning models, it’s imperative that they continuously check the models to make sure they are working as intended and that the data is valid, unbiased and not corrupted in any way.

Deepchecks, an Israeli startup has come up with a solution to continuously test models from development through production. Today, the company announced a $14 million seed investment and the general availability of the open source version of the solution. The company also has a commercial product, Deepchecks Hub, that it began testing earlier this year.

CEO Philip Tannor and CTO Shir Chorev spent time in the Israeli military where they worked on machine learning. Chorev said they learned firsthand about the power and limitations of the technology, and that’s why they decided to start a company to work on quality-checking models.

“So both Philip and I were both in positions of leading AI research and bringing models to production. And we both saw the huge value potential that the algorithms in AI in general have, but alongside that we also noticed that they do face some quite significant challenges, specifically being able to actually do what they’re supposed to be doing and doing it properly,” Chorev told TechCrunch.

She says it’s hard to know what’s going on inside these models without tools like Deepchecks, which help data scientists make sure the data they are using are not biased, misinformed or based on corrupt input data – and generally performing well over time.

She and Tannor, being practitioners, realized that what was lacking was a tool to help continuously check these models throughout the process to test and validate different aspects of the model like the data’s integrity. So like any good entrepreneurs, they built the tool they wanted and open sourced it. More recently they also built a commercial version of the product, one that provides standard enterprise features like strong security and authentication.

“We brought quite a lot of understanding and hands-on experience in the field and also knowing the problem and what we would have wanted,” she said. Those experiences were relevant and helpful as they built the company. The company reports the open source product has been downloaded over 500,000 times and is being used by the likes of AWS, Booking.com and Wix, as well as regulated companies they can’t name publicly.

The company launched in 2020 and took about a year and a half to launch the first version of the open source tool.

Chorev recognizes that she is a relative rarity as a female co-founder, but she says she has always been interested in fields dominated by males and that has only driven her throughout her career. “Personally, I do feel that from a young age that this wasn’t something that was going to stop me. Instead, it’s like okay, it’s harder, so I’m gonna push and get it,” she said.

“I mean, it’s a challenging route, any way you choose it, but I do try to be accessible to help others to encourage people, whether they’re females or males, to take this path if that’s what they want.”

Today, the company has about 15 employees with plans to grow that number by 25-30% this year. She is concerned about building a diverse workforce, even while recognizing it can be a challenge for a young company. “It is a big challenge and we do put the effort in order to encourage diversity now, and also going forward as we grow,” she said.

Today’s funding round is really the combination of two tranches of money, a $4.4 million pre-seed round when they launched and $9.5 million, which closed at the end of last year. The investment was led by Alpha Wave Ventures with participation from Hetz Ventures and Grove Ventures.

Deepchecks snags $14M seed to continuously validate ML models by Ron Miller originally published on TechCrunch