Arrikto, a startup looking to accelerate the machine learning development cycle by enabling engineers and data scientists to treat data like code, has come out of the shadows today and announces a $ 10 million Series A round. The round was chaired by Unusual Ventures, with Unusual's John Vrionis joining the board.
“Our technology at Arrikto helps companies overcome the complexities of implementing and managing machine learning applications, ”said Constantinos Venetsanopoulos, CEO and co-founder of Arrikto. “We're making it very easy to set up end-to-end machine learning pipelines. In particular, we make it easy to create, train, and deploy ML models in production using Kubernetes and intelligently manage all the data around it intelligently. "
Like so many developer-centric platforms these days, Arrikto is all about "left shifting". Currently, the team says, machine learning teams and development teams don't speak the same language and use different tools to build models and bring them to production.
"Similar to how DevOps has shifted deployment to the left on developers in the software development lifecycle, Arrikto is shifting deployment to the left on data scientists in the machine learning lifecycle," said Venetsanopoulos.
Arrikto also aims to remove the technical barriers that still make machine learning implementation difficult for most organizations. Venetsanopoulos noted that just as Kubernetes has shown companies what simple and scalable infrastructure can be, Arrikto can show them what a simpler ML production pipeline can look like – in a Kubernetes-native way.
The heart of Arrikto is Kubeflow, the google – Incubated open source machine learning toolkit for Kubernetes – and in many ways, you can think of Arrikto as an enterprise-ready version of Kubeflow. Among other things, the team built MiniKF to run Kubeflow on a laptop and used Kale, which allows engineers to create Kubeflow pipelines from their JupyterLab notebooks.
As Venetsanopoulos noted, Arrikto's technology does three things: It simplifies the deployment and management of Kubeflow, enables data scientists to manage it with the tools they are already familiar with, and creates a portable environment for data science, which involves data versioning and sharing between Teams and clouds enabled.
While Arrikto has stayed off the radar since it launched in Athens, Greece in 2015, Venetsanopoulos' founding team and CTO Vangelis Koukis have already managed to get a number of large companies to adopt its platform. Arrikto currently has more than 100 customers, and while the company is not yet allowed to name any of them, Venetsanopoulos said they include one of the largest oil and gas companies, for example.
And while you might not consider Athens a startup hub, Venetsanopoulos argues that this is changing and that there is plenty of talent there (although the company is also using the funds to grow its sales and marketing team in Silicon Valley). "There is first-class talent from first-class universities who have not yet been developed. It's like we have an unfair advantage," he said.
"We see a strong market opportunity as companies seek cloud-native solutions to take advantage of machine learning," said Vrionis of Unusual. "Arrikto has followed an innovative and holistic approach to MLOps across the entire data, model and code lifecycle. Data scientists can reduce time to market through increased automation and collaboration without the need for engineering teams. "