Private AI for engineering and architecture
Twenty years of projects, reports and tender documents on the practice's server. Every new bid closely resembles an old one, but finding it costs more than rewriting it.
A technical consultancy's real asset is its project history, and it is almost never searchable. Private AI indexes reports, tender documents, annexes and site records, and answers citing the specific project and section. It runs on a server in the practice, so documentation covered by an NDA or by the client's intellectual property is never disclosed to an outside provider.
The real pains
The ones that show up exactly when there's a deadline.
Rewriting what is already written
The report for a similar project exists, but finding it in the archive costs more than drafting the section again.
Tenders against the clock
Hundreds of pages of requirements and a week to respond. Pinning down which ones apply burns half a day of a senior engineer.
The technical criteria are unwritten
How the practice resolves a recurring detail is known by three people, and they always have to be interrupted.
NDAs that block the obvious tool
The team would use cloud AI, but client contracts forbid passing project documentation to third parties.
What AI does about those four
Reuses your own history
"Building services reports for industrial buildings using this solution" retrieves the sections of earlier projects that cover exactly that, with the document cited.
Turns a tender document into answers
Ask about eligibility requirements, deadlines or scoring criteria and get the answer with the paragraph quoted.
Makes the practice's criteria searchable
Construction details, internal notes and documented decisions stop living only in the seniors' heads.
Fits the NDAs you already signed
With nothing leaving the practice, there is no third-party disclosure to authorise and no confidentiality clause at risk.
The project isn't yours to disclose
In engineering and architecture the problem isn't only data protection: it's contractual. Much of a project's documentation is covered by confidentiality agreements and by the client's intellectual property. Uploading it to a cloud tool is disclosure to a third party, enterprise plan or not.
There, private AI isn't a preference. It's the only thing that fits: the model runs on a server in the practice, nothing reaches the internet, and you can answer in writing when a client asks where their documentation ends up.
What it does not do
It does not calculate, model or draw: it replaces neither calculation software nor CAD or BIM, and it does not interpret the geometry of a drawing. It signs nothing and takes on no technical responsibility: review by a qualified engineer or architect remains mandatory. And it knows no standard you never indexed: it works over your own archive, not over an external body of regulations. And it produces no exhaustive listings ("give me every project using this solution"): it answers over the documents it retrieves for each question, not by sweeping the whole collection.
Before you ask
Can it run calculations or produce drawings?
No. It does not replace your calculation software, CAD or BIM. It is a documentation tool: it finds, compares and drafts from projects, reports and standards you have already written.
What about the NDAs we sign with clients?
That is the main reason this sector asks. Many contracts forbid passing project documentation to third parties without authorisation, and uploading it to a cloud AI is, literally, disclosure to a third party. With the server in the practice there is no disclosure to justify.
Does it read drawings and technical PDFs?
It reads text: PDFs, Word documents and Excel sheets; reports, tender documents, annexes and site records. It does not interpret the geometry of a drawing; that is what your technical tools are for.
Is it useful for preparing tender submissions?
It is one of the clearest returns: it retrieves the equivalent sections from your own previous submissions and prepares a draft the team reviews, instead of starting from a blank page each time.
Do we have to reorganise the project archive first?
You don't need to tidy it up: it indexes the folder structure exactly as it is, because the mess is precisely the problem it solves. What does get reviewed at the start is where that archive lives and how it is set up, because that determines how it connects.
Let's start with the next tender
A 30-minute call to see which part of the project archive is worth indexing first.