Technical Example: A Source-Linked Knowledge Assistant

Short answer

This technical example describes an architecture in which employees search company documents within their permissions and get answers that cite sources. Permissions are applied at retrieval, the assistant states when information is insufficient if no suitable source exists, and instructions inside a document are not applied as system authority. It is a design example, not a client case.

Example use case

In this architecture example, staff search company documents within their access permissions and receive source-linked answers. The assistant makes the supporting document and information visible together.

Data flow

Collect documents and permissions → index with source and version identifiers → authenticate the user → search authorised sources → generate an answer with references. State information gaps when no suitable source exists.

Freshness and access

Remove index records when sources are deleted. Apply permission changes to search. Document instructions must not become system authority. Responses should have verifiable sources and document versions.

The limits of this example

The example is not the result of a specific client project and contains no measurement results such as an accuracy rate. In a real implementation, quality should be measured with your own documents and evaluation questions.

Evaluation

Test correct sources, old versions, conflicting documents, unanswerable questions and unauthorised access. Assess source correctness and access behaviour separately from fluent wording.

Frequently asked questions

Two questions we are often asked on this topic.

Which document types is this architecture suited to?

Text-bearing sources such as PDF, Word and wikis. For scanned documents, text-extraction quality should be measured first.

What happens when a source is deleted?

Its index record is removed too, and the deleted document is not used to produce answers. Update and deletion flows are part of the design.