AWS has introduced a method for building a semantic layer for agentic AI, leveraging Stardog’s Semantic AI Application and Amazon Bedrock AgentCore. This approach allows agents to query data sources like Amazon Aurora and Amazon Redshift to answer customer 360 questions without requiring extract, transform, and load (ETL) processes. The semantic layer is designed to provide AI agents with a unified view of enterprise data, enabling them to reason over fragmented data sources with the same fluency as human analysts. The solution also supports integration with AWS compute services such as Amazon EKS, Amazon ECS, and AWS Lambda, with AgentCore simplifying the deployment by bundling authentication, hosting, and tool credentials into a single managed service. This development aims to address the long-standing challenge of enabling AI agents to work with enterprise data that is scattered across multiple systems with inconsistent definitions. The semantic layer uses an ontology-driven approach, translating queries into SQL at runtime and allowing agents to compose answers from various data sources. By combining foundation models with a semantic layer, AWS is positioning itself to support the next generation of agentic analytics that can operate on live data without human intervention. Source: awsml
The semantic layer complements traditional Retrieval Augmented Generation (RAG) techniques by providing AI agents with a structured view of enterprise data, enabling them to answer analytical questions that require joining records across systems and applying business rules consistently. Unlike RAG, which relies on text-based retrieval, the semantic layer translates queries into SQL and joins data from multiple sources at runtime, ensuring that the results are accurate and aligned with business definitions. This approach allows AI agents to operate on live data without needing pre-built models or datasets, reducing the dependency on data engineers and analysts. The semantic layer also supports access control through named graphs, which are labeled subsets of the knowledge graph that can be accessed based on user roles. Virtual graphs, which are not stored in Stardog but fetched from external systems like Aurora and Redshift, enable the layer to work with a wide range of data sources while maintaining data integrity and governance. By integrating Stardog’s federated knowledge graph with Amazon Bedrock AgentCore, AWS is offering a scalable solution for enterprises looking to deploy agentic AI that can reason over complex, fragmented data. Source: awsml
Enterprise analytics has long struggled with the challenge of delivering timely and accurate answers to business questions, with traditional methods like scheduled reports and dashboards failing to meet the demands of self-service analytics. Even self-service BI tools require data engineers to build the right models for specific queries, leaving human analysts as the bottleneck for unstructured or unexpected data. Generative AI agents represent the next evolution in this space, as they can reason over data, plan queries, evaluate results, and iterate on the fly, without waiting for human input. However, the success of these agents depends on their ability to access and understand enterprise data, which is often scattered across systems with conflicting definitions. The semantic layer addresses this by capturing business context and metrics once, allowing all agents and tools to reuse this information. This approach ensures that AI agents can provide reliable, business-context-aware answers while maintaining data consistency and governance. Source: awsml
