Build a semantic layer for agentic AI on AWS with AtScale and Amazon Bedrock AgentCore
aws ·
This article details the third installment in a series on building semantic layers for agentic AI on AWS, specifically utilizing AtScale and Amazon Bedrock AgentCore. It demonstrates how to implement a Customer 360 pattern using AtScale’s dimensional modeling approach, which differs from the graph-based or virtualization methods used in previous posts. The solution employs the Semantic Modeling Language (SML) to define datasets, hierarchies, and measures, versioned in Git and exposed via SQL, XMLA, and Model Context Protocol (MCP). A Strands agent hosted on AgentCore uses Claude Sonnet 4.6 to interpret user questions and generate queries against the AtScale MCP server. The architecture ensures secure identity propagation through AgentCore Identity, enforcing role-based access control and row security at the semantic layer before queries reach Amazon Redshift. AtScale’s autonomous aggregate management optimizes performance by detecting query patterns and routing them to pre-materialized summary tables. Finally, the post outlines deployment on Amazon EKS and verifies that the semantic layer provides consistent, governed answers across both AI agents and traditional BI tools.
aws