Large Scale On-Demand Low-Latency Near Real-Time Predictions
summary
Sony PlayStation builds an internal relevance-scoring platform that serves personalized recommendations under 20 milliseconds at scale — about 50,000 requests per second — using Aerospike for hot features, S3/PARQUET for bulk data, Spark/Flink for streaming, and Kubernetes for inference. Models retrain daily or weekly, updates reach the online store within ~5 seconds, and monitoring plus A/B tests validate behavior; on-demand scoring replaces nightly full rescoring and improves freshness for long‑tail users.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 0:54LowLatency
- 1:17Playstation
- 2:57Sony PlayStation 5
- 5:38Relevance Scoring
- 5:54Other Platforms
- 6:27ML Pipeline
- 7:48Customer Item Intelligence
- 8:53Feature Store
- 9:08Relevance Platform
- 9:49Recommender Systems
- 11:17Challenges
- 11:29Topics
- 12:07ML Engineer
- 12:30Data Science
- 14:29Feature Generation
- 15:38Training Data
- 18:01ML Skills
- 19:18Questions
related talks
15mPowering PlayStation personalization to millions
33mPowering Real-time Recommendations with Machine Learning on GCP
41mBig Data Serving: The Last Frontier. Processing and Inference at Scale in Real Time
20mDataEngBytes 2023 - PER-T1-06 - Kamal Hossain
19mBuild Real time, AI powered Apps
12mQuickUpdate: a Real-Time Personalization System for Large-Scale Recommendation Models
45mReal-time, Streaming Advanced Analytics....
29mMachine Learning in Production: What to do after you’ve trained the models?