Kepler AutoML Targets Next-Gen Business Analysts
As more companies roll out digital infrastructure, they are ingesting greater volumes of data that can be used by business analysts to gauge customer intent and boost transactions. Complexity and lack of data scientists have made that transition harder for mid-size firms looking to monetize “dark” data.
Machine learning vendors are therefore automating key aspects of data science workflows that would allow domain experts to customize pipelines and algorithms based on specific data types. AutoML approaches are promoted as boosting the quantity and quality of machine learning models produced on, say, a monthly basis.
That’s among the goals of a new AutoML platform unveiled this week by Stradigi AI. The Kepler platform simultaneously seeks to address the shortage of data scientists, the resulting inability to move AI models to production and then scale up those models. The strategy focuses on freeing domain experts to select the data science tools needed to get machine models out the door faster, with an initial focus on “high-value” use cases like inventory control or customer churn that are most likely to yield actionable results.
The “sweet spot” for data science automation are mid-size enterprises seeking to reap the rewards of digital transformation, said Per Nyberg, Stradigi’s chief commercial officer. Hence, Kepler automates data science steps to quickly move business analysts up the machine learning curve.
The strategy involves helping domain experts identify high-value use cases like predicting customer intent and “just get them going,” added Nyberg, former vice president of AI at Cray Inc.
Nyberg said in an interview that Kepler is also positioned as a productivity tool for data scientists currently struggling to build and scale machine learning models. Montreal-based Stradigi estimates that about half of AI adopters struggle to move models to production while three-quarters are unable to scale models once deployed. Hence, Nyberg said, Kepler aims to train machine leaning-savvy “business analysts of the future.”
Along with retail, media and advertisers, Stradigi is also focusing on mid-sized manufacturers along with transport and logistics firms. The strategy behind Kepler involves buttressing investments in business intelligence tools to add predictive analytics for use cases like demand forecasting or maintenance.
The company said its deep learning workflows leverage more data formats to expand the number of potential use cases. To that end, the automation platform ingests tabular, text and image data, then automatically generates pipelines after data cleansing, which the company said is the most labor-intensive step in the workflow. Depending on data type, Nyberg said users can select the best pipeline and algorithms to accelerate training, inference and transfer of models to production. Once in production, Kepler also generates API code to connect ML inference to BI systems.
Kepler is initially available on Amazon Web Services with availability on Microsoft Azure coming soon. Depending on the use case, the automated data science workflow supports GPUs for HPC use cases, Nyberg added.
Early customers include ReclameAQUI, a popular Brazilian web site initially using Kepler to gauge customer churn and intent prediction.
Stradigi AI, which earlier focused on software engineering and mobile app development, launched the Kepler project in 2014 as machine learning models were slowly beginning to move into production with mixed results. Those early struggles led to the rise of AutoML tools that would free data scientists to focus on data analytics rather than underlying plumbing. Kepler aims to take that notion a step further by automating the development of deep learning models for domain experts.