AI Briefing
KOSign in

GS Energy Builds Enterprise Agent Data Platform with SageMaker Unified Studio and Bedrock AgentCore

·2026.10.06 15:33

Key point

Data integration across three subsidiaries including GS Power enables approximately 690 users to access data via catalog subscriptions

1 / 16

Details

GS Energy has built an enterprise platform connecting 'Data-to-Agent' by combining Amazon SageMaker Unified Studio as a data catalog and governance hub with Amazon Bedrock AgentCore as an agent runtime. This platform integrates distributed data from power generation subsidiaries and enables business users to find and utilize data interactively.

Architecture and Data Governance

The platform structures the data lake with a Medallion architecture consisting of Raw, Conformed, and Curated layers, applying access control at the table, row, and column levels via Amazon Athena and AWS Lake Formation. Data usage scope is defined based on the project and subscription models of SageMaker Unified Studio, managing permissions so that only approved data is accessible to agents and business users. Currently, data from three subsidiaries—GS Power, GS EPS, and Incheon Energy—is connected, with approximately 690 users utilizing the platform.

Agent Execution and Security

Agent execution is handled by Amazon Bedrock AgentCore Runtime, ensuring security and stability through per-session microVM isolation. User authentication integrates with the internal SSO Keycloak, while AWS service calls adopt a dual authentication structure using separate IAM credentials. Agents verify permissions via MCP (Model Context Protocol) servers to ensure data access is restricted to the calling user's permission scope. Currently, the catalog lookup agent is in operation, while the Text-to-Any agent supporting natural language-based data analysis is in the development and deployment stage.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.