Multi-tenant LLM analytics with row-level security: How we built a secure agent on AWS

This article describes how PAR Technology built a production-ready multi-tenant LLM analytics system that enforces row-level security through three independent security layers.

• Layer 1: Cryptographic request signing with AWS SigV4 verifies user identity before any processing occurs
• Layer 2: Semantic validation via Amazon Bedrock reasoning engine validates questions against supported metrics before data access
• Layer 3: Split-Plane SQL architecture programmatically generates pre-filtered data sandboxes scoped to each user's authorized data
• LLM operates only on pre-filtered schemas, preventing access to unauthorized data even if compromised or manipulated
• System processes 50,000+ queries with zero cross-tenant data exposure incidents in production
• Each layer operates independently; data boundaries enforced deterministically at architecture level, not reliant on LLM compliance

The three-layer approach ensures multi-tenant analytics systems maintain security boundaries deterministically while enabling business users to query data in natural language.