Wallaroo Scales for the Future With Replicated

Wallaroo outline

While the dystopian world of Blade Runner or the bright futurism of The Jetsons has yet to come to fruition, a common thread that ties our collective predictions of the future remains true to this day. At some point, machines will learn to think on their own.

This shared prognostication is best realized today with the emergence and embrace of machine learning (ML) - the practice of training data models to imitate how human beings learn. ML has empowered data science to reach new heights and enjoyed widespread adoption among enterprises to add business value. But now enterprises have found that finding patterns in the data is the easy part. What’s hard is taking machine learning models out of the dev environment and operationalizing them into the business against live data and feeding their downstream business applications.

This is why Wallaroo was founded - to facilitate the last-mile of the enterprise machine learning journey. Wallaroo enables enterprises to get ML into their production environments to impact the bottom line with incredible speed and efficiency. Their platform provides a simple, secure, and scalable way to deploy, run, and observe ML in an enterprise production environment.

Wallaroo consists of a single platform with three core components:

  • A self-service toolkit for data scientists and ML engineers to deploy, manage, and collaborate on machine learning models in production.
  • A blazingly-fast compute engine for breakthrough scalability and performance.
  • And advanced observability and experimentation framework to provide model insights and drive results.

The Problem

The problem, however, is that the modern enterprise data ecosystem consists of various platforms and applications living in different environments. For example, a financial services company might do model training in the cloud but have to deploy on-prem due to regulatory reasons. Or some enterprises might be all-in on a certain cloud but then acquire a new company whose legacy IT is on a completely different system.

One of Wallaroo’s core differentiators was to allow customers to continue using their preferred model development tools and then deploy into any destination environment. Wallaroo however didn't have a simple way of deploying this software to customer-controlled environments without an extreme amount of support from their internal teams. As a seed-stage company, Wallaroo knew that to properly grow and scale, they would have to figure out a way to operationalize the delivery of their software to a multitude of customer environments.

As Younes Amar, Head of Product at Wallaroo, recalls, "We needed to simplify the process of installing Wallaroo into different customer environments. It was important to allow our customers, data scientists and ML engineers, to realize the value of our product without dealing with the complexity of installing it or integrating it in their environment."

Another critical concern for Wallaroo was building a community edition of their platform that data scientists could use for free while evaluating if the platform was robust enough for their demands. A community edition would go a long way towards building trustability among their prospective users and demonstrate the value that Wallaroo’s platform delivered.

This, however, would have to wait. Wallaroo couldn't devote valuable resources towards a community edition until they had a simple, scalable solution for deployment and support.

A cloud-based architecture solution would allow Wallaroo to distribute their software to the major cloud platforms, and thus reach a broader addressable market, as well as operationalize product updates with a centralized repository for customers to access releases. Additionally, the architectural shift would remove the complexities of delivering updates individually and the hand-holding required to get customers up and running.

Wallaroo had looked into building a cloud-based solution in-house but had quickly ascertained that the project would take at least a year to complete. This was unacceptable for a seed-stage company where engineering resources are vital to the success of its core platform. To sideline important engineering assets on supporting installations would set Wallaroo back significantly when it came to growing their market and scaling out their platform, and enhancing their core differentiating features like advanced model observability and hyper-efficient runtime engine.

Not to mention their planned community edition, which would have to wait in the wings for now while they solved the issues with delivering and supporting their commercial offering.

The Solution

Wallaroo's search for a solution led them to Replicated, a Kubernetes application delivery and management platform. Replicated provides commercial and open source tools to streamline the distribution and management of 3rd-party applications to complex enterprise environments and all major cloud platforms.

First on the docket was figuring out a way to deploy Wallaroo's platform to any environment that their customers requested. Replicated delivered in spades on that front.

The first significant benefit that Wallaroo noticed after partnering with Replicated was an accelerated time to market. Wallaroo had estimated that getting a cloud-based solution out to customers would have taken them a year's worth of work to build in-house. With Replicated, that time was drastically reduced.

"Based on our own estimations, it would have taken us a year to be fully operational with a cloud-based solution if we built it ourselves. With Replicated, we were able to be up and running in less than a month," recalls Younes. "Having Replicated take care of all the complexities, management, and licensing helped us be focused solely on delivering our product."

Replicated's partnership allowed for an extraordinarily augmented approach to getting their platform into customer environments. It also granted Wallaroo's engineering team the ability to speed up product development by removing the need to focus on a cloud-based installer.

"The way I would quantify the return on investment with Replicated was that we could double our product development velocity without running into scale inefficiencies or problems when it comes to delivering our products to customers," says Younes.

And those customers that requested Wallaroo on different cloud platforms? They could rest easy knowing that Wallaroo could deliver wherever they needed. To add to that, Wallaroo even saw expanded benefits from the Replicated installer's abstraction layer between cloud services.

As Younes puts it, "We have the ability with this cloud offering to be up and running in all three clouds, which makes it easier to install and integrate for most of our customers. At the same time, from a product maintenance perspective, we know that we don't have to worry about cog-related issues. If any issue comes our way, we know what to trace it back to and how to resolve it efficiently."

Finally, Wallaroo could focus their attention back on offering a community edition to build trustability and foster a sense of transparency and openness to the AI/ML community. Thankfully, they were able to leverage Replicated to assist in this endeavor.

Using Replicated's Community Licensing features, Wallaroo can assign as many community licenses as they want through the Replicated vendor portal. These licenses are locked behind a limit usage and allow users to try out the Wallaroo platform while evaluating if it's right for their needs.

Younes praises the benefits of managing both paid and community licensing within one, simple platform. "The best part was that we didn't have to develop a new licensing or product mechanism. We leveraged the Replicated community licensing offering to essentially be able to deliver two versions of the same product simultaneously without any additional effort."

When users wish to upgrade to the full commercial product, Wallaroo simply has to change the license types through the vendor portal's UI and convert them to paying customers. This showcases Wallaroo's transparency with the community. It also allows them to target new revenue streams by offering a nearly full-featured edition of their products to data scientists worldwide.

The Future

With their delivery, management, and licensing needs accounted for, Wallaroo could refocus on scaling their organization for the future. They quickly began engaging with investors and were able to successfully raise a $25M Series A round, thanks in part to the solutions put in place for self-hosted customer installs through Replicated.

Our investors were really impressed that we had figured out how to manage installations at scale,” says Younes. “A large part of the feedback we received on our Series A round was how we got ahead of a major scale problem by figuring out a managed installation solution with Replicated.

When asked if he would advocate for others to partner with Replicated, Younes replies, “I would definitely recommend Replicated to any organization that has a product they need to be up and running in cloud environments or on-premises.

Please schedule a demo today to learn more about how Replicated can empower your enterprise software.