Ask-your-documents internal knowledge search
Years of proposals, reports and SOPs become searchable in prose, with citations and with permissions respected by construction.
This is a reference design we build from — one of a set that are not delivered client projects, and the figures are estimates rather than measured results.
Trigger
A question typed in Slack/Teams or an internal web page
Timeline
1 wk scoping and evaluation set → 3-4 wks build → 2 wks pilot with one team
Stack
Vertex AI Search or pgvector with embeddings on Cloud Run · Drive / SharePoint connectors with ACL propagation · Gemini or Claude for synthesis · Slack/Teams bot · Retrieval-quality evaluation set · Per-query cost telemetry
Outcome
Institutional memory becomes queryable. The 'who knows about X' tax falls. Permissions are respected by construction rather than by policy.
The process today
People ask colleagues questions the company has already answered, because years of proposals, reports and SOPs are unsearchable.
What the automation does
Indexes the approved document set with permission awareness, so a person only ever sees what they already have access to. Answers in prose with citations and links to the source file and page. Returns “I don’t have that” rather than inventing an answer. Logs unanswered questions as a content gap report for the knowledge owner.
Where a human stays in the loop
The knowledge owner reviews the content gap report and decides what to document next. Anyone can follow a citation back to the source page instead of trusting the answer blindly.
What you end up owning
The retrieval index, the connector and ACL-propagation configuration, the evaluation set, and the cost telemetry — all inspectable and running in your own environment.