FanDuel, the premier online gaming company in North America, recently hosted an event at its New York City office to discuss the challenges of implementing production machine learning (ML) systems. In attendance were engineering and data science experts and enthusiasts who were eager to learn from FanDuel’s first-hand experience in this area.
During the event, Tim Gestson, Lead Engineer of FanDuel’s platform team, and Morgan Hsu, Director of Engineering – Data and Machine Learning at FanDuel, discussed why and how they adopted Tecton’s feature platform to power their ML models and data-driven applications. In this post, I’ll share some of the highlights from the discussion.
The journey to a production ML platform
Tim and Morgan shared FanDuel’s journey in designing its production ML platform, Omni, and emphasized the importance of the technological choices they had to make along the way, as well as the people and process considerations they had to take into account.
They also shared the various tools the company used to build out its ML capabilities, including Databricks, Delta Lake, Unity Catalog, and Tecton. Gestson and Hsu highlighted the importance of choosing the right tools and frameworks to build an effective production ML platform that meets precise technical requirements and aligns with the organization’s culture and values.
The need for a feature platform
To take their ML capabilities to the next level, FanDuel needed a feature platform that could model data specifically for ML use cases, provide managed solutions to simplify operations, and address some of the challenges the teams faced with managing data backfills and orchestrating the backfilling process when a feature changed. Finally, they wanted a tool that provided both real-time and batch functionality and also seamlessly integrate into their existing stack.
The team chose to explore managed feature platforms rather than build their own because taking the home-grown feature solution route would have required significant resources, expertise, and long-term maintenance. Additionally, a managed platform would prove more cost-effective for FanDuel since they could focus on developing and deploying ML models for their business needs instead of building and maintaining the underlying infrastructure to support a homegrown solution.
After evaluating different options, the teams decided to go with Tecton’s managed feature platform because it allows them to govern and integrate with FanDuel’s existing tooling while providing online and offline functionality for real-time serving and generating reliable batch training sets. Furthermore, Tecton fit nicely into FanDuel’s Databricks deployment and provided online and offline functionality, allowing real-time features to be materialized into DynamoDB tables with a real-time API, while Spark could still be used for offline training.
Finally, according to Tim, one of the most important aspects of integrating Tecton into their own system is that the feature platform provides a way for different teams to contribute to an inter-source repo to develop features, resulting in a growing library of features powered by Tecton. This means that engineers, data scientists, and ML engineers can all work across teams, together or off one another, to build and improve features for their models.
Omni (and Tecton’s) impact on FanDuel’s ML strategy
Integrating Tecton’s feature platform with FanDuel’s Omni ML platform has enabled personalization efforts and a responsible gaming model, which is a crucial use case for the company. The model relies on ~71 features that are processed and backfilled using Tecton. With Tecton’s help, FanDuel processes hundreds of millions of rows of data and scores millions of results.
With Tecton’s feature platform integration, FanDuel has streamlined its feature management process and developed more accurate models that have a positive impact on its business. As the company continues to grow and explore new opportunities in the world of sports entertainment and gaming, its partnership with Tecton will remain crucial in enabling its teams to innovate and maintain its competitive edge.
During their talk, Tim and Morgan demonstrated the importance of choosing the right tools and frameworks for building production ML systems. Their experience showcases the potential of feature platforms like Tecton to support the development and deployment of ML models, ultimately driving business growth and enhancing the customer experience.
Interested in seeing how Tecton can help you drive more business growth or improve customer experience? Sign up for a personalized demo today.