The Key Role Of Feature Stores In Deploying Real-time AI
summary
Real-time AI requires 50–200 ms end-to-end latency and feature stores that serve features in under 10 ms (ideally <1 ms); Redis is shown as the online store in multiple production pipelines, with a cited fraud use case at ~50 ms. Benchmarks claim P99 <1.5 ms and Redis Enterprise is 3× faster and 14× cheaper than DynamoDB. Switching Redis strings to hashes cuts latency ~40% and cost ~5×. Offline training commonly uses Snowflake or BigQuery and online serving pairs Feast or Tecton with Redis.
AI-written summary · about
chapters (13)
- 0:00Introduction to real-time AI
- 1:50Redis in AI and ML
- 3:15Feature stores explained
- 4:41What real-time means
- 7:33Feature store data sources
- 8:37Redis as online feature store
- 10:27Why Redis is fast
- 12:16Redis editions and scaling
- 14:06Offline vs online feature stores
- 15:10Redis in production use cases
- 15:53Redis beyond feature stores
- 16:56Redis deployment options
- 17:39Learning resources and Q&A
related talks
43mThe Challenges of Deploying Real-time AI for Finance and how Open Source can help
19mML feature store a cornerstone in modern data architecture for AI/ML
11mBuilding Real-Time Feature Pipelines
25mBuilding A Feature Store For Hyper Growth
19mBuild Real time, AI powered Apps
14m5 great ways to reimagine your data layer
08mReal-Time Data Systems: Speed, Scale, and Resilience
44mFeature store