How Octave built a secure multi-tenant AI foundation on Amazon Bedrock
aws ·
Octave Reliance developed a secure, multi-tenant AI foundation on Amazon Bedrock to unlock institutional memory for regulated industries without compromising data isolation. The team prioritized building a horizontal platform over individual features to address four key tensions: strict tenant separation, isolated agent execution, untrusted input handling, and high-throughput embedding of two billion documents. By leveraging Cohere Embed v4 for multilingual and multimodal retrieval, the system achieved significant storage efficiency while maintaining accuracy across diverse document types. Security was enforced through Amazon Bedrock AgentCore, which utilizes dedicated microVMs and per-user identity propagation to prevent cross-tenant data leakage. Additionally, Amazon Bedrock Guardrails provided defense-in-depth against prompt injection and sensitive information disclosure by filtering inputs and outputs at multiple stages. The resulting Complaint Triage Assistant reduced investigation times from hours to minutes by automatically assembling relevant historical context for quality handlers. This architectural approach cut deployment time for new AI use cases by 50 percent, allowing Octave to deliver semantic search capabilities ahead of schedule.
aws