Scalable Recommendation Systems on Ray

Build large scale real time recommendation systems with the Ray ML libraries.

Build the future with scalable ML platforms

Recommendation systems are a common core capability of any organization with a large product or service catalog. Small measurable improvements in matching products to users can have a large impact on sales and an improved user experience. This makes ROI easy to measure and investment justifiable if the development cost and time to market is low enough.

Established methods like collaborative filtering can now be augmented by deep learning (DL), reinforcement learning (RL), and contextual bandits. Organizations find using these techniques is often difficult due to the limitations of their currently deployed machine learning systems.

The flexibility of Ray allows customers to both migrate their existing recommender systems, as well as augment them with these newer techniques.

The right way to scale for Recommendation Systems

Train, test, deploy, serve, and monitor machine learning models efficiently and with speed with Ray and Anyscale.

  • Best-in-breed algorithms

    Use the latest recommendation system algorithms from contextual bandits to transformer-based deep learning models to traditional collaborative filtering techniques. The Ray ecosystem natively supports RLLib, and integrates with most Python based ML libraries including SciKit learn, XGBoost, PyTorch and Tensorflow.

  • Easily compose multiple ML models and logic

    Getting the best results from a recommendation system requires combining multiple models and business logic, while being able to meet the latency needs of a live application. Organization must be able to easily combine models and logic into an execution graph that executes in a highly scalable cluster.

  • Scale training across large sets of data

    Recommendation system models are built from large scale data sets of users, products and interaction events. Organizations need to be able to pipeline their data for training and tuning in parallel, allowing training times to dramatically decrease while improving model performance.

Relevant offers delivered at exactly the right time

Wildlife Studios serves in-game offers three times faster with Ray Serve

Already using open source Ray?

Experience the power of infinite scale and eliminate complex production operations.