Building an MLOps platform at HelloFresh

Speaker: Erik Widman, Ph.D., Director of Product for Machine Learning @ HelloFresh

Abstract: Deploying machine learning models to production can be complicated. At HelloFresh, our data science teams have grown organically, solving problems across the company using different models, from marketing to supply chain applications. Because of this rapid growth, we suffer from inconsistent MLOps tooling, and some teams struggle to build pipelines. These challenges lead to model scaling difficulties across geographies, pipeline reliability issues, and tech debt.

To solve these challenges, we have developed an MLOps platform that will work for most models at HelloFresh, reducing the time to create a pipeline from weeks to hours. In addition, the pipeline is reliable, scalable, easy to use, and standardizes infrastructure across HelloFresh.

This presentation will cover our product strategy for MLOps, technical execution for the platform, and strategic considerations for data products.

Speaker Biography: Dr. Erik Widman is the Director of Product for Machine Learning at HelloFresh. Previously he has built AI innovation teams at Accenture and consulted for F100 companies on a variety of products within the fields of Computer Vision, NLP, Topic Modeling, Voice AI, classification, and forecasting. Before his work in AI, he spent ten years in the biotech industry designing Neurostimulators and image processing algorithms for ultrasound machines.

Outside work, he enjoys spending time with his wife, Anna-Mi, and his 15-month-old son Julian. He is also a musician in the band Love in October and is passionate about cooking and exercise science.