Octave Reliance has described an AI foundation for its cloud-based quality-management platform, which serves organizations in regulated industries. The company says the system is designed to search more than 2 billion documents, including complaints, corrective and preventive actions, audit findings, scanned forms, images and email attachments, without crossing customer tenant boundaries.

The AWS blog post says Octave chose to build a shared foundation before developing individual AI features. Its first application is a Complaint Triage Assistant. Workshops with AWS also surfaced other potential applications, including complaint summarization, audit-readiness monitoring, incident narrative generation and predictive quality-risk scoring.

The foundation uses Cohere Embed v4 on Amazon Bedrock to create embeddings and Amazon OpenSearch Service for per-tenant vector stores. Octave says Embed v4’s multimodal and multilingual capabilities allow text and images to be represented in one vector space without separate translation or image-processing pipelines. Testing with representative production queries led the team to select 512-dimensional embeddings, which it says reduced storage by about 66% compared with the model’s 1,536-dimensional default while maintaining target accuracy.

For execution, the architecture uses Amazon Bedrock AgentCore Gateway and isolated microVM sessions. The source says authenticated user identities, temporary credentials and tool-level authorization are carried through the execution chain. Amazon Bedrock Guardrails are applied at input and output stages. Octave reports that the foundation has reduced deployment time per use case by 50% and that the team committed to onboarding its largest tenant within a single working day.