Skip to main content
Datenschaftler

Solution pattern · not a customer claim

Knowledge assistant with controlled data and permissions.

The pattern shows how internal documents can be made securely searchable, permissions can remain intact, and cited answers can support an existing workflow.

Starting point

Knowledge is distributed and governed by different permissions.

Policies, manuals, and case documents often live in several repositories. A useful solution must respect currency, metadata, and existing permissions.

System boundary

The assistant answers. It does not decide.

The pattern supports research and summarisation. Business decisions, approvals, and exceptions remain with accountable staff.

Architecture

Four layers with clear accountability.

Each layer has its own quality and operational requirements. This makes it possible to distinguish a data, retrieval, model, or process issue.

Information flow
  1. 01

    Sources & access

    Approved documents, versions, and user permissions are carried forward.

  2. 02

    Data pipeline

    Content is extracted, structured, checked, and indexed traceably.

  3. 03

    Search & answer

    Combined search returns sources; the answer remains grounded in the retrieved context.

  4. 04

    Use & operation

    Feedback, quality measurement, logging, and escalation support daily operation.

Guardrails

Reliability comes from the system as a whole.

A capable model cannot replace missing data ownership or unclear process boundaries.

  1. 01

    Permissions remain intact

    Users only see content they are already permitted to access in source systems.

  2. 02

    Sources remain visible

    Answers reference the documents used and support business review.

  3. 03

    Quality is measured

    Real questions, expected sources, and business criteria form a versioned test set.

  4. 04

    Failure paths are designed

    Uncertain answers, missing sources, and technical failures lead to clear alternatives.

Possible technology components

  • Azure AI Search
  • Azure OpenAI
  • Microsoft Entra ID
  • Azure Blob Storage
  • Application Insights

Your system context

Which sources and workflow need to be connected?

We assess whether this pattern fits or a simpler architecture would be more appropriate.