Open-source MLOps vendor aims to give users a ‘sneak peak’ into AI deployments

ClearML cofounder and CEO Moses Guttmann knew there was potential for his firm’s open-source based MLOps tool. What he didn’t know was that the emergence of ChatGPT in late 2022 would accelerate the whole market and his company along with it.

Today, ClearML announced a series of updates to its platform, alongside strong growth in the first quarter of 2023 with more than 1,300 global enterprise companies now using the ClearML MLOps platform. The momentum is being bolstered by growing demand and interest in machine learning (ML) model development and deployment, as organizations of all sizes look to benefit from the technology.

With MLOps, the basic idea is to provide organizations with the tools needed to help manage the workflow for building and testing machine learning. ClearML has both an open-source project as well as an enterprise edition that debuted back in September 2022.

Among the new capabilities that ClearML is launching is a feature that the company calls ‘sneak peak’ that goes a bit beyond traditional MLOps functionality. With sneak peak, users can iteratively deploy and preview test models in real time, while models are still in development. ClearML is also adding in new model lineage capabilities that can help with AI explainability.

“We’ve seen 150,000 data scientists use ClearML just in the last quarter,” Guttmann told VentureBeat. “We attribute a lot of interest to the ChatGPT hype, with basically everyone understanding that they really have to get onboard.”

A ‘sneak peak’ into the future of model development

The MLOps workflow typically involves a set of steps to help data scientists build a model.

What ClearML is doing with its sneak peak approach is allowing data scientists to easily deploy internal machine learning–backed applications for the product and business units to experience as part of the development process. The goal, according to Guttmann, is to make ML development more accessible and to shorten the time it takes organizations to get value out of the whole process.

“ClearML before this was more targeted toward a machine learning engineer or developer audience,” Guttmann said. “With sneak peek we’re also targeting the product people.”

An emerging use case that Guttmann has seen is the implementation of ML directly inside of products with a continuous learning approach. He noted that there are organizations now using ClearML where models are constantly being trained as data is being collected.

“We’ve seen companies deploy machine learning automations as part of the product itself,” he said. “So the product itself has this capability of training itself.”

Improving AI explainability with model lineage

Another area of improvement for ClearML is with the addition of new model lineage capabilities.

With model lineage, an organization can track where different elements of a model come from and how they change over time.

“As time goes by, it’s very important to be able to do some forensics on models being deployed,” Guttmann said. “So if something goes wrong, we can trace back on the originating codebase and data that was used to train that specific model.”

With model lineage, he said ClearML now provides clear visualizations to help understand who created a model and where the model is being used in production. Being able to track lineage is a critical element of AI explainability, helping organizations to be able to demonstrably track what has gone into model development.

“We are trying to advocate safe and secure model development,” he said.