US LBM, a building materials distributor, partnered with 7Rivers to deploy Snowflake Cortex Agents integrated with Microsoft Teams, enabling business users to query enterprise data in natural language. The architecture uses OAuth and SCIM to sync EntraID identities to Snowflake roles, enforces Row-Level Security so users only see authorized data, and leverages Semantic Views to give the AI business-domain understanding. Specialized agents (rather than one generalist) are recommended to avoid conflicting instructions. An automated evaluation pipeline using 'Golden Records' and a Judge LLM scores responses on accuracy, completeness, concision, and security. Monitoring is handled via Snowflake Query History, Snowsight dashboards, and a custom audit view. The result is a self-service BI chatbot that reduces ad hoc query backlogs while maintaining strict governance.
Table of contents
Crushing the Ad Hoc Query Backlog: How US LBM Democratized Data with Snowflake Cortex and MS TeamsThe Core Challenge: Making Data Accessible yet SecureThe Architecture: Connecting Teams to SnowflakeThe FlowSecurity Deep Dive: Identity and Row-Level FilteringThe Brains: Cortex Agents, Cortex Analyst, and Semantic Views and MaintenanceOperationalizing the Semantic LayerCortex AgentGet Dimitry Borochin’s stories in your inboxStrategy: Specialized vs. Generalist AgentsTrust but Verify: Automated EvaluationThe “Judge” LLMMonitoring and Observability for the Cortex AgentBuilt-in Snowflake ObservabilitySimplified Custom MonitoringConclusion: A New Era for Data AccessReady to build?Sort: