DataRobot’s Zepl acquisition bridges the AI divide

DataRobot revealed yesterday during an online AI Experience Worldwide conference that it has acquired Zepl as part of a larger effort to enable data scientists to customize AI models developed on its platform. Zepl created an open source Apache Zepyl notebook that enables data scientists to collaboratively develop and analyze code written in Python, R, or Scala.

DataRobot also unveiled a slew of updates, including an ability to clone, edit, and reconfigure machine learning blueprints created using the AutoML framework at the core of its platform for constructing AI models. That Composable ML feature enables organizations to integrate their own custom training code with the AutoML framework employed by DataRobot, senior VP Nenshad Bardoliwalla said. “Data scientists can now tweak our AI models,” he added.

DataRobot is also adding a Continuous AI feature that enables organizations to implement a policy that defines when an AI model should be retrained based on the level of drift detected by the MLOps governance framework embedded in the DataRobot platform. Alternatively, organizations can simply decide to schedule an AI model to be retained at a given interval.

There is also now a No Code AI App Builder that automatically converts any model into an AI application without requiring coding. Widgets, data visualizations, and prebuilt templates enable AI applications to be constructed in a few minutes.

Finally, DataRobot has added a grading tool to generate an automatic scorecard grading based on an evaluation of data quality, robustness, accuracy, and fairness. The company is making available a Bias & Fairness Production Monitoring tool that monitors models for bias after they have been deployed in a production environment. Previously, bias detection could only be applied to AI models as they were being developed.

Democratize AI

DataRobot is driving an effort to democratize AI using a framework that automates most of the routine tasks associated with aggregating data and then training an AI model. So far, business analysts and executives have been the primary users of a DataRobot platform accessed via graphical tools. A challenge  DataRobot has encountered is that many AI models today are built by data scientists using notebooks and other types of open source toolkits. The acquisition of Zepl sets the stage for enabling data scientists to employ the same platform as end users and business analysts to automate the training of an AI model, Zepl CEO Dan Maloney said.

Zepl claims its notebook has been downloaded more than 500,000 times. This provides the foundation for AI model training to become a more collaborative effort involving end users, business analysts, data scientists, and developers that is closer to a team sport, Maloney added. “They can use any open source toolkit,” he said.

Developers can invoke AI models running on the DataRobot platform via a REST application programming interface (API). Alternatively, they can embed the server running an AI model as a Java Archive (JAR) file or in a Docker container within their application.

The pace at which organizations want to employ AI to automate a wide range of digital processes is far outstripping the available supply of data scientists. As a result, there is a chronic need to enable end users and business analysts to create AI models without the aid of a data scientist. However, there are still complex AI models that need to be custom built by a team of data scientists. The challenge organizations now face is recognizing the difference between the level of skills required to build one AI model versus another.