Making Company Knowledge Accessible

The theory is set. Now let's get practical: How do you make your real company knowledge — distributed across Confluence, SharePoint, emails, PDFs, and databases — accessible to AI?

Step 1: Inventory Knowledge Sources

Create an overview of all knowledge sources:

Step 2: Build Connectors

Confluence

• Atlassian REST API for page content and metadata
• CQL (Confluence Query Language) for targeted extraction
• Webhooks for incremental updates

SharePoint / OneDrive

• Microsoft Graph API
• Delta queries for incremental syncs
• Carry over permission filters (important for compliance!)

Emails

• IMAP/Exchange connector
• Only internal emails and threads with business relevance
• PII detection and masking before ingestion

Databases

• SQL views as defined interfaces
• Change Data Capture for real-time updates
• Embed schema documentation as additional context

Step 3: Data Pipeline

Source → Extraction → Cleaning → PII Filter → Chunking → Embedding → Vector DB
           ↓                                                              ↓
        Scheduler                                                    Metadata Store
     (daily/weekly)                                               (source, date, permissions)

Step 4: Access Control

Critical: The RAG pipeline must only return information that the requesting user is authorized to see.

• Carry over permissions from source systems
• Filter against user roles at query time
• Regular audits of access patterns

Common Pitfalls

• ❌ Migrating everything at once (instead of iterating)
• ❌ Ignoring permissions
• ❌ Not flagging outdated documents
• ❌ No feedback loop with users

Practical tip: Start with a single source (e.g., Confluence) and 10 power users. Collect feedback, optimize, then scale to more sources. A RAG system thrives on iterative improvement.