Real-time personalized recommendations with Amazon SageMaker and Amazon-managed Valkey

This article describes an architecture for delivering real-time personalized product recommendations using Amazon SageMaker and Amazon-managed Valkey, addressing the challenges of serving relevant results at scale.

• SageMaker hosts sentence transformer models to convert customer queries into semantic vectors
• Valkey provides sub-10ms vector similarity search using HNSW indexing on millions of products
• Architecture handles latency, throughput, semantic understanding, and personalization requirements
• Optional enrichment from DynamoDB layers additional signals like preferences and trending items
• Parallel processing and configurable pipelines keep end-to-end latency under 100 milliseconds
• Valkey chosen for in-memory performance, rich filtering, and multiple indexes per cluster
• Index populated via batch embedding generation and pipelined HSET commands

The solution demonstrates how to combine semantic embeddings, vector search, and optional enrichment to build production-scale recommendation systems that feel instant and personalized.