There are numerous observability platforms that monitor and evaluate cloud software, like Dynatrace and ServiceNow, which flag potential code errors or failures so engineers can find and fix them. Arize AI says it is bringing that same approach to AI models and applications.
Arize is an AI observability platform that helps companies evaluate their AI products as they are building them and then monitor those products for errors and issues once they are up and running. Arize’s platform works with a variety of AI applications, from machine learning and computer vision to generative AI.
Jason Lopatecki, Arize co-founder and CEO (pictured above left), told TechCrunch that Arize uses a “council of judges” approach to monitor and evaluate AI. This approach includes evaluating AI with different AI models — which Lopatecki joked is, yes, very meta — in addition to having humans in the loop.
The idea behind Arize came from Lopatecki’s previous company TubeMogul, a brand advertising company, which was acquired by Adobe for over $500 million in 2016.
Everything at TubeMogul ran on AI, Lopatecki said, and when it would break it would be a “big deal” since the technology was so complicated. Aparna Dhinakaran, a co-founder and CPO at Arize (pictured above right), who met Lopatecki through TubeMogul, had run into similar issues developing language models without having the proper tools to test and evaluate as she built.
“We both saw the problem space and really had that idea that AI is going to be high stakes in more and more organizations everywhere,” Lopatecki said. “It’s so complicated, it’s really hard to tell what it’s doing, when it’s broken and how to fix it.”
The pair launched Arize in 2019 with an initial focus on the AI trend of the day: predictive machine learning. Lopatecki said that when Arize got started, it was really just an idea. Today, five years later, the market gets the problem and Arize’s platform works with everything from AI agents to generative AI.
“So the last two years have been, I would say, explosive, explosive in growth,” Lopatecki said. “Simply because [AI] is more accessible. Everyone’s a prompt engineer. Every engineer is a prompt engineer. Everyone is integrating [AI] products into their product lines.”
Arize now works with enterprises including Uber, Klaviyo, and Tripadvisor, among others. The company also has an open source offering, Arize Phoenix, which has more than two million monthly downloads.
The Berkeley, California-based company recently raised a $70 million Series C round led by Adams Street Partners with participation from M12, Sinewave Ventures, and OMERS Ventures, among other investors, in addition to strategic backers including Datadog and PagerDuty. This brings the company’s total funding to more than $130 million to date.
The company plans to put its latest round of funding toward improving its main product and doubling down on growing AI segments, including voice and AI agents. Dhinakaran joked that while their open-source product may be their biggest competitor, the company plans to put more money into developing that product, too.
“Our open source Phoenix has just been growing, it’s been growing massively, and so I think we love that. We love open source,” Dhinakaran said.
The AI observability and evaluation space is becoming increasingly crowded. Dhinakaran said that they think that Arize offers both pre- and post-launch evaluations, and can be used across a variety of different AI applications, which helps the company stand out; although, there are companies with very similar offerings like Galileo, which has raised $68 million in venture funding, and Patronus AI, which has raised $20 million in funding.
“It’s so hard to build the [infrastructure] to do this, right?” Lopatecki said. “It’s kind of why I think the Microsofts and Datadogs are investing in us, or making a bet on us. I think people also now see how big this market can be. You’re going to have a lot of little guys. You’re gonna have big people jumping in it, and I expect it to be a fast, growing, large market.”
Add comment