AWS and Stardog Architecture for Enterprise Agent Data Analysis

AWS ML Blog · rss · 2026-07-10

This article introduces how to use the Stardog Knowledge Graph and Amazon Bedrock AgentCore to build an enterprise-level agent analysis system. It solves data silos and semantic inconsistencies across data sources (like Aurora and Redshift) without needing ETL data movement.

The article notes that while current LLMs are strong at planning and generating SQL, the bottleneck in enterprise data analysis is the lack of unified business context. Letting models directly access fragmented underlying data easily leads to queries that are syntactically correct but logically flawed. To fix this, the author proposes a three-layer architecture for building agents:

Model Layer: Handles natural language understanding and planning (e.g., using Anthropic Claude), but the model itself doesn't understand the business.
Semantic Layer: Uses an ontology-driven knowledge graph to capture business concepts, rules, and entity relationships. It dynamically translates user queries into SQL for underlying multi-source data and performs federated queries on shared identifiers.
Agent Runtime Layer: Uses AgentCore to host agents, unifying authentication, hosting, and tool credential management.

The article further compares the use cases of the semantic layer versus RAG (Retrieval-Augmented Generation): RAG is great for retrieving text, while the semantic layer is better for analysis tasks requiring cross-system real-time record linking and complex business rules. They are often used complementarily in production. Finally, the article provides two integration paths to connect agents to Stardog: using SPARQL tools directly, or via a Model Context Protocol (MCP) server as a gateway tool.

Original post →

AWS and Stardog Architecture for Enterprise Agent Data Analysis

More from coding & agent