Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn how LinkedIn transformed their recommendation system from slow, offline training to real-time, continuous model updates using Ray in this 29-minute conference talk from Ray Summit 2025. Discover the limitations of traditional offline recommender workflows where model freshness lagged by days or weeks, leading to outdated recommendations and missed opportunities to adapt to changing user behavior. Explore LinkedIn's migration to online training with Ray, which created a tight feedback loop for continuously refreshed models, resulting in improved recommendation quality, higher click-through rates, and up to 8.9× cost savings through incremental retraining. Examine the production system design including high-throughput streaming data generation, scalable Kafka ingestion pipelines, and a nearline training loop powered by Ray for fast iterative updates. Understand key optimizations for throughput, latency, and system stability, along with lessons learned from deploying Ray in mission-critical production environments and essential architectural choices for reliability and scalability. Gain actionable insights for building Ray-powered online training systems that maintain model freshness, reduce costs, and drive measurable business impact in real-time recommendation scenarios.
