Exclusive Interview with Vid Jain, CEO & Founder, Wallaroo

Machine learning models take a lot of time and huge swathes of data to yield results. A typical machine learning cycle consumes most of the resources for data gathering and wrangling, which may or may not be responsible for generating business value eventually. It is only the last mile execution, i.e., deploying an ML model into the market and observing them for necessary changes in accordance with changing market, that helps make ML models profitable. Having observed this trend, Wallaroo Labs, an MLOps platform enables companies to leverage the power of live data to deploy their ML models successfully. Analytics Insight has engaged in an exclusive interview with Vid Jain, CEO & Founder, of Wallaroo Labs.

1. Kindly brief us about the company, its specialization, and the services that your company offers.

Wallaroo Labs is an MLOps platform focused on the last mile of machine learning (ML) implementation, i.e., getting ML into a production environment. By saying the last mile, we mean more the part of deploying an already built model. We make testing, deploying, running, managing, and observing these models in production easy and highly performant. We're privately held and backed by several leading venture capital firms including Microsoft's M12, Boldstart Ventures, and other leading enterprise software VCs.

2. With what mission and objectives, the company was set up? In short, tell us about your journey since the inception of the company.

We started in 2017 as a group of engineers dedicated to solving the increasingly common problem of analyzing large amounts of data via computational algorithms efficiently and at scale. By applying our expertise in building distributed computing systems in industries such as high-frequency financial trading and AdTech, we built a high-performance compute engine like nothing else on the market. While our customers could now efficiently analyze their data and use it to run ML models at scale, they had one biggest challenge: bringing those models online easily and understanding how they were performing to generate business value sustainably. Like most organizations, they followed the standard DevOps playbook: data scientists would build ML models to solve a business problem, and engineers would launch them using a patchwork of open-source software and containerized model approaches. However, they found that the standard DevOps playbook just didn't work for ML. Models often had to be painstakingly re-engineered, meaning the deployment software couldn't process data fast enough — even when running on an alarming amount of computing resources — and it was unnecessarily difficult to see how models were performing to measure their ongoing accuracy. We started Wallaroo with the mission of making it the easiest, fastest, and best-performing way for enterprises to generate sustainable business value from their AI/ML programs.

3. Brief us about the proactive Founder/CEO of the company and his/her contributions to the company and the industry.

Vid Jain comes from an academic background, with a Ph.D. in Theoretical Physics from UC Berkeley University. He helped build the algorithmic trading desk at Merrill Lynch, where he first stumbled upon the "P&L of ML" concept. That is, this is where he noticed that the return on investment into your AI and ML program wasn't from data wrangling or building models in a lab – but from having those models in production, driving results. As those models can expire quickly as the markets changed, you need an easy way to continually monitor their ongoing performance and make it easy to deploy, test, and undeploy models.

4. Tell us how your company is contributing to the IoT / AI / Big Data Analytics / Robotics / Self-Driving Vehicles / Cloud Computing industry of the nation and how the company is benefiting the clients.

Data scientists are building better and more complex models in the lab, but the lab environment is very different from having a model running at the edge. For example, you can spin up cloud computing to train and run a model on a large quantity of batch data in the lab. But when you have that model deployed at the edge, for example, a computer vision model trained to detect pedestrians requires you to run that model with ultra-low latency at the edge, where millisecond latency matters. This means there is very limited computing available to run the model at the edge. Our purpose-built engine makes it possible to still run that model at high performance using much less computing. There is also the issue of model testing and observability. How do you test different models across a fleet of smart devices? Most conventional MLOps solutions don't make testing and managing model versions across hundreds or thousands of devices easy. We think of ourselves as an "ML control center," giving you easy visibility and management of all your models in production and testing on the cloud, on-premise, or at the edge.

5. Kindly share your point of view on the current scenario of Big Data Analytics and its future.

At the beginning of the Big Data gold rush, the promise was that data would be this incredibly valuable asset, so the important thing was to hoard as much as possible and figure out how to use it later. Digitization was generating more data than ever, so enterprises thought if they could just get a handle on this stream of direct customer data, it would fix everything. So, cloud technologies came along and other SaaS providers made data engineering easy and cheap. But data on its own doesn't do much. Data science became so hot that you could take all this data from various sources and find predictive patterns in that data. Voila! AI and ML models!! This is where I think most enterprises are now. Even the digital laggards are hiring data scientists and building models for different parts of the business. But they're still not generating a return on their investment. What's missing? Data scientists are great at looking at business problems and building models, but that's not necessarily the same as making production-ready code. So, you have your AI leaders, maybe the top 5% of companies, that have figured out how to take these AI models built in R&D and then take them into production to generate value. But you now have these early and late adopters coming up behind them who may not have the resources or ability to hire an army of ML engineers to get these models from R&D into production. We're very excited about this next huge wave of enterprises that have made the investments and now are ready to make AI an integral part of their operations if they can figure out how to make a scalable, agile AI program.

6. Please brief us about the products/services/solutions you provide to your customers and how they get value out of it.

Our product has three main components: a self-service toolkit for easy model deployment and management; a distributed compute engine purpose-built for ML to infer faster, using less compute; and observability, insights, and dashboards to monitor the ongoing performance of models in testing and production. Together these three components make applied AI programs scalable in the enterprise by allowing you to deploy and undeploy models in seconds without major reengineering, making it economically feasible to run these models (often, compute costs can be greater than gains from ML models if not using a specialized engine), and making it simple to test and monitor for data and model drift.