Skip to content
All use cases

Automation

Knowledge portal over your own documents

One search box over SharePoint, handbooks and wiki. Answer with a source link or an honest "no idea".

New joiners ask the same questions for three months that are already answered in a document nobody can find. An internal search that knows which page it does not know is worth more than a chatbot that guesses.

The situation

A 70-person SMB has SharePoint, a Confluence wiki, a four-year-old handbook in PDF, and a handful of procedures sitting as Word files in a folder tree on the file server. On top of that, Teams channels carry messages where somebody once typed the right answer.

The HR person gets the same questions twice a week: how is leave booked, how much travel cost is reimbursed, has the work-from-home rule changed. The answer is somewhere. But "somewhere" is unfindable if you do not know where to look.

The temptation is to drop a generic chatbot on SharePoint. The trouble: it hallucinates with conviction. An employee asking about leave days and getting a fabricated number back creates more work than it saves.

What we did

We build an internal search-and-answer layer that only speaks about documents it has actually seen. No open internet, no inventions from model memory.

Phase 1: index. We hook SharePoint, Confluence, and the PDF handbooks into a vector index. Every chunk is stored with source and version. Updates run automatically.

Phase 2: answer layer. A question is answered with citations from one to three documents, with a link to the source page, and with a date. When the search results are too weak (low relevance score, conflicting sources) the system says it does not know and forwards the question to HR.

Phase 3: feedback. HR sees which questions come in often and which documents are missing. Documentation effort goes there, rather than into free-form forum writing.

What it delivered

Repeat questions drop out of the HR mailbox. What stays is the kind of question that actually needs a personal answer.

The portal flags which documents are stale, because the questions about them often come back with low confidence. That is a cheap way to start a documentation maintenance cycle.

For new joiners: onboarding has one place to look first. That removes a small bit of friction in looking something up instead of interrupting a colleague.

What this wasn't

Not a generic ChatGPT bot on top of your SharePoint. It knows what it does not know, and does not pretend.