“MLOps,” a compound of “machine learning” and DevOps, is a set of practices that aims to maintain machine learning models in production both reliably and efficiently. It primarily involves testing models in isolated systems and then transitioning them into production, but MLOps applies to the entire lifecycle — from integrating with model generation, orchestration, and deployment to governance and business metrics.
According to a 2021 Deloitte report, MLOps is set to significantly shorten the time it takes to put a newly developed model into production. That’s caught the attention of enterprises looking to adopt AI across their organizations. Based on research from Cognilytica, the MLOps market could expand to nearly $4 billion by 2025.
The growing list of startups developing MLOps solutions includes Comet, Iterative.ai, ZenML, Landing AI, Domino Data Lab, and Weights and Biases. Another — Qwak — recently joined the fray with a platform that automates MLOps processes and allows companies to manage models the moment that they’re integrated with their products. Qwak today announced that it raised $15 million in a funding round co-led by Leaders Fund and StageOne Ventures, with participation from Amiti Ventures and individual investors.
Qwak was founded in 2021 by Alon Lev, Yuval Fernbach, Lior Penso, and Ran Romano. Lev was the former VP of data at Payoneer, while Fernbach was a machine learning specialist at Amazon Web Services. Penso led business development management at IronSource, and Romano previously headed up machine learning engineering at Wix.com.
“[W]e saw that sophisticated enterprises have the resources and knowledge to build in-house MLOps tools that enable them to adopt and run machine learning applications in production,” Lev told VentureBeat via email. “However, most companies in the world do not have the assets required, and this is the most significant barrier to machine learning adoption. Qwak’s vision is to build a machine learning engineering platform that, through automation, enables all companies to adopt machine learning to build and maintain machine learning-driven products at scale — without having to build all the plumbing themselves.”
Qwak aims to standardize machine learning project structures with code, versions data, and parameters — supporting the deployment of models with metrics, logging, and alerting capabilities. The platform also offers a data lake, a centralized repository that allows customers to store all structured and unstructured data, as well as a feature store for engineering features. (Feature engineering is the process that takes raw data and transforms it into features that can be used to create a model.)
Qwak also has automation features that allow customers to configure triggers based on different model “layers,” infrastructure, data, and statistics. They also allow users to run actions — whether they’re “Qwak internal,” like triggering a build and deployment, or external, like calling external APIs and integrating with third-party apps.
“[I]n the last year, we have seen an increasing amount of cases where customers want to implement a model per each one of their customers — mainly business-to-business businesses. Previously, this would have required very complicated and costly infrastructure to support it, but Qwak enables ‘model per customer’ architecture that not only optimizes the infrastructure, but also automates the operational work that is required to manage many models per use case,” Lev continued.
A growing market
As my colleague, Louise Columbus, writes, enterprises have an urgent need to get more machine learning models into production. That’s because 87% of data science projects never make it to completion. Delays getting machine learning models into production are symptoms of larger, more complex problems — including lack of production-ready data, integrated development environments, and more consistent model management. According to IDC, 28% of all AI and machine learning projects fail because of these factors.
MLOps platforms are by no stretch a silver bullet. Companies adopting them are often stymied by unrealistic expectations, data discrepancies, a lack of communication, long chains of approvals, among other issues. Qwak’s proposition, then, is that its platform can alleviate the bottleneck that exists between the development and data science teams more cost-effectively than rival platforms.
“In the last decade, many companies have hired teams of data scientists without achieving the business goals they expected. Now, it’s on these teams to prove their impact on the bottom line — and Qwak helps them deliver value on their machine learning investment,” Lev said. “C-suite managers understand the value of machine learning, but aren’t able to deliver on it because of the operational challenges they face in building machine learning based products … Qwak empowers companies to successfully implement machine learning in their products, scale the number of models in production and iterations on those models, and does so without having to hire more machine learning talent.
With the capital infusion, Qwak — which has 21 employees across its R&D center in Israel and its San Francisco, California office — plans to invest in product development and expand its sales and marketing departments. The company claims to already have customers in Yotpo, Guesty, Skyline AI, and JLL.