MLOps, or DevOps for those working with machine learning models, has seen a boom of interest in the last year, and that should come as no surprise: organizations want to bring more machine learning to bear in their data science stacks, but that requires building and training models, cleaning up data and making sure they work as they should. Today, a startup called Striveworks building MLOps tools to handle that work is announcing $33 million in funding.
This is the startup’s first-ever outside funding, and the round getting closed now underscores both the surge of interest in the wider area of artificial intelligence, but also Striveworks’ own traction within that, with the company’s ARR growing 300% annually over the last two years.
The $33 million is coming from a single investor, Centana Growth Partners, and Striveworks, based in Austin, plans to use it for hiring and for further product and business development. The funding comes as what is often described as an opportunistic round: Striveworks has been in business for five years, run as a “capital efficient” startup that made profit and invested that back into its growth, according to Jim Rebesco — the CEO who co-founded the company with Craig Desjardins, Eric Korman and Tony Manganiello.
Rebesco did not disclose current customer names but said that they span a range of verticals that include government and the financial sector that use machine learning to build services or run their businesses, “highly regulated industries and national security applications and associated areas like computer vision intentionality, satellite imagery and commercial imagery,” he added. The company also has partnerships with AWS and Azure to work on data in those clouds. (Notably, it currently does not have a similar partnership with Google.)
The problems that the company is tackling are things that Rebesco — a neuroscience PhD who previously had a long stint at financial services company Virtu — said he and his co-founders regularly encountered at previous companies, which Striveworks essentially aims to combat with realism.
It starts, he said, with what he described as the “day-one problem” of how to build appropriate machine learning models to fit one’s objectives. But that’s in some ways the easy part. The complexities start really after that.
“Does it do what you expect it to do, and when you put it into production, does it continue to perform as you expect?” he said. “We focus on what happens next.”
Rebesco describes himself as a “failed physicist” (a reference to his pre-PhD work, I think), who learned an important lesson about AI models: They are all statistical and therefore bound to bring up failures. “So one of the key elements of responsibility is not just knowing there will be errors but putting an automated and thoughtful plan in place to address that.”
He believes that this is something that has to be increasingly considered as the use of AI becomes more ubiquitous. “Data models, AI and ML models, are increasingly important not as ephemeral models. Whether it’s credit scoring or healthcare, those databases are being stored, and are being queried. But how do you query [effectively] is so much wrong?”
The company aims to tackle this by way of its flagship platform called Chariot, which can be used to help prepare data, build models and then run those models in production. Using a low-code format that is geared towards teams collaborating, features on the platform include model-in-the-loop annotation, the ability to import models and use previously catalogued data models (from your own organization), the ability to build custom workflows, query the “provenance” of data in your sets, and the ability to integrate third-party tools, among other features.
There are now a lot of startups (and bigger companies) in the market working MLOps solutions — a few that we’ve covered include Seldon, Galileo, Aries, and Tecton. Bigger systems integrators are also getting in on the action, with McKinsey recently acquiring Iguazio.
Ben Cukier, the Centana partner who led the investment, said that Striveworks had a clear advantage over these, in that the business itself is being run very well, a sign both of the operations of the company and what they’re achieving.
“They are at the scale where their growth rate, in the triple digits, is where most are when they are only at Series D. I got a look at their really efficient use of capital and was blown away. In 27 years of investing, I’ve only seen a couple of companies able to achieve that kind of scale without outside capital. It’s a rare occurrence. These are real customers, with seven-figure contracts, and with net retention numbers that would be the envy of a lot of other companies.”
The company is not disclosing its valuation but Cukier described the current market not as “quiet” but simply “normal” — that is, back to business as usual after what have been several very heady years.
Striveworks snaps up first funding of $33M to build tools for machine learning operations by Ingrid Lunden originally published on TechCrunch