Skip to content

AI for engineering documents

29docs applies generative AI to engineering archives. An agent in Microsoft Teams answers technical questions in plain language, based on the EDMS vault and the network folders, and cites the document and revision behind each answer.

Short answer

AI for engineering documents is an agent that reads a company's technical archive, from the EDMS vault to network folders, and answers questions in plain language, citing the source document. 29docs builds the agent with Azure AI Search and RAG, inside Microsoft Teams, in the client's environment and within the permissions already in place.

How it works

  1. Sources

    Network folders and the EDMS vault: Meridian, ProjectWise or SharePoint.

  2. Data in order

    Normalisation, a data dictionary and metadata modelled for the AI.

  3. Indexing

    Azure AI Search indexes with each document’s text and metadata.

  4. Agent in Teams

    Answers with RAG, always citing the document and the revision.

Where technical search loses time

The right document almost always exists. The problem is finding it. The data sheet is in the EDMS vault, the previous revision sits in a network folder and the vendor’s comment is in an email. To answer a simple question, such as a line’s design pressure, an engineer opens three systems and checks which revision applies. In a pilot, that time fell from 20 to 3 minutes per search.

What AI does on an engineering archive

  • Answers technical questions in plain language, with the source document and revision.
  • Finds the current version of a drawing, data sheet or procedure.
  • Extracts metadata from legacy documents that entered the vault without attributes.
  • Summarises what changed between two revisions of a document.

Security and governance

The agent runs in the client’s tenant and infrastructure. Whoever can access a document keeps that access, and nobody gains new permissions because of the AI. In the pilot, access is by invitation, in a dedicated test team in Teams. Client data does not train public models.

Why AI starts with the data

AI answers well when the archive is in order. Consistent metadata, standard numbering and a master document register that says which revision applies are what separate a correct answer from a plausible one. It is the work 29docs has done for years in engineering document management, which is why every AI project starts with an archive assessment.

Updated on .

Getting started: a 4 to 6 week pilot

The pilot runs in your environment, on a set of documents from the vault and the network folders, and ends with what you need to decide on the full rollout.

  1. Archive assessment
  2. Normalisation and data dictionary
  3. Indexing
  4. Agent in Teams
  5. User validation
  6. Network folder study
  7. Go/no-go report

What you get

  • Documented user validation
  • Infrastructure, software and licences for the full rollout
  • Sizing study of the network folders
  • Total cost of phase 2

20 → 3 min per search, measured in a pilot.

Talk to 29docs

Frequently asked questions

Does it work with Meridian, ProjectWise and SharePoint?

Yes. The agent indexes the Accruent Meridian vault, the Bentley ProjectWise datasource, SharePoint libraries and network folders. Each source comes in with its metadata and revisions.

Do I need to migrate the archive before using AI?

No. The agent reads the archive where it is. What is usually needed is putting the data in order: normalising metadata, reviewing numbering and flagging the current revision. If the archive is moving platforms, plan the EDMS migration alongside.

What about scanned documents?

Scanned documents go through OCR before indexing so their text becomes searchable. Answer quality depends on scan quality, and the pilot shows this with real documents.

Who sees what?

Each person only gets answers from documents they can already open. The agent follows the EDMS and network folder permissions, runs in the client's environment, and client data does not train public models.

Ready for your next audit?

A 30-minute conversation. No pitch. We want to understand your document reality before talking about any solution.

We reply within 1 business day. Your data is never shared.