Build a semantic layer for agentic AI on AWS with Stardog and Amazon Bedrock AgentCore

This article demonstrates how to build a semantic layer using Stardog's knowledge graph to enable AI agents on Amazon Bedrock to query federated data across Aurora and Redshift without ETL.

• Semantic layers provide governed, business-context-aware access to enterprise data across multiple systems using ontologies and mappings
• Stardog federates Aurora and Amazon Redshift through virtual graphs that mint shared entity identifiers (IRIs) enabling cross-warehouse joins without physical integration
• Named-graph security controls role-based access to sensitive data like PII, with the same query returning different results for different users
• Reasoning rules encode business definitions like "Big Spender" once in the ontology, eliminating duplication across queries and dashboards
• Amazon Bedrock AgentCore provides managed hosting, authentication, and credential management for production agents querying the semantic layer
• Two integration paths: direct SPARQL tool (Path A) or Stardog Cloud MCP server with Voicebox natural-language layer (Path B)
• Customer 360 use case demonstrates joining operational customer data from Aurora with analytics purchase data from Redshift through shared customer identifiers

Semantic layers complement RAG by enabling reliable agentic analytics through governed access to live enterprise data with consistent business definitions across all sources.