1. Introduction
The AI Document feature transforms the Celiveo 365 infrastructure into an intelligent Corporate Memory system known as AI-Doc. This feature captures Documents submitted to the Celiveo 365 document capture infrastructure, indexes their content, and securely stores them in a siloed database.
It integrates with Microsoft Universal Print and Microsoft Teams and it enables authorized users to query the Corporate Memory to retrieve information, provide concise summaries and analysis from processed documents using natural language in their preferred language. It also provides citation of sources, and metadata regarding the Document’s origin.
Access to this data is strictly governed by a Zero Trust architecture, ensuring users only retrieve information relevant to their assigned cohorts.
2. Architecture
The solution architecture is designed to securely capture document output workflows with AI-driven processing, analysis and retrieval in a Zero Trust model. All components are automatically deployed by celiveo to the organization’s Microsoft Azure tenant, ensuring data remains private, encrypted (AES-256 at rest and in transit), and isolated by design.
AI Document High-Level Architecture Overview
- Source: Microsoft Universal Print Queues configured to capture document to AI Document (Documents submitted by users)
- User context: (UPN + assigned Cohorts) captured at submission time. Default Cohort assignment can be configured.
- Text/content extraction
- Indexing for semantic search and retrieval
- Metadata enrichment (filename, submitter UPN, timestamp, page-level info)
- Encrypted storage in Azure (tenant-isolated blob storage)
- Data strictly siloed using Cohort tags inherited from submitting user, user group and Entra ID
- Zero Trust enforcement: documents only visible to users sharing ≥1 matching Cohort
- Documents stored with AES-256 encryption + per-document unique keys
- Natural language queries in user’s preferred language
- Authentication via Microsoft Entra ID → scoped access to user’s Cohorts only
- Semantic search + generative AI response generation
- Response includes: concise summary, source citations (filename, page, author), metadata
- Commands for session control, notifications, document listing & deletion
Key Security Principles
- Zero Trust: Every access decision based on identity + Cohort matching
- Data Isolation: No cross-cohort leakage (e.g. HR documents stay HR-only)
- No external model leakage: Processing and generative AI occur within tenant boundary
- Auditability: Origin, submitter and access traceable via metadata
Post your comment on this topic.