Skip to content
ProQu
All use cases
AI SolutionsIndustrial company, EUExample

Search over company knowledge

Manuals, contracts and policies are spread across drives and an old intranet. We build a system in the client’s cloud that answers from the current document, cites the page and respects the access rights people already have.

The problem

The current version of a policy or a manual is hard to find. Files sit in shared drives, in mail and in an old intranet, so people ask colleagues instead.

A chatbot pointed at “all the files” would make this worse. It would answer from a draft, from a superseded contract or from a folder the person is not allowed to read. The job is larger than one agent: it needs ingestion, access control, citations and a way to know an answer is wrong before anyone relies on it.

What we build

A production system in the client’s cloud.

  • Ingest. Documents are collected from the sources the company already trusts. Each file keeps its source, its date and its access list.
  • Prepare. Text is extracted, split and indexed. A superseded document is marked, and the search prefers the current one.
  • Ask. A person asks in plain language. The answer cites the document and the place in it. If the sources disagree, the answer says so.
  • Access. A person sees only what they could already open in the source system.
  • Evaluate. Before launch, a set of real questions from the staff is scored: did the answer cite the right source? The set is kept and run again when the index or the model changes.
  • Hand over. Documentation, a runbook and a named owner on the client side.

How it runs

  • The system is deployed in the client’s cloud account.
  • Models are called through the client’s provider agreement, or an open model is hosted in their network when they require that.
  • Logs show the question, the sources used and the answer. Retention follows the client’s policy.
  • A person reviews an answer before it is sent outside the company or used in a compliance decision.

What the client gets

  • A production system, with documentation.
  • Citations on every answer, and a visible signal when the source is old.
  • An evaluation set built from their own questions.
  • Handover or ongoing support, as agreed.

The result

Staff ask a question and get an answer with the document and the location. They can tell when a source is out of date, and access matches the permissions they already had.

Have a similar problem?

Tell us about it. We reply within one working day.

Talk to us