Engineering Data AI
Engineering data AI gives your engineers answers from the data your company already owns: PDM records, ERP article and cost data, wiki pages, documents on file servers and CAD geometry, with the source next to every answer. Your systems stay where they are. The AI reads from them, writes nothing back and replaces none of them.
Engineering data AI: your systems, one read-only layer, two kinds of answers
- PDM or PLMcommercial, or self-built on a SQL database
- ERPfor example SAP S/4HANA, SAP Business One or proAlpha
- Wiki and intranetfor example Confluence or SharePoint
- File serversspecifications, technical PDFs, drawings
- CAD geometrySiemens NX, CATIA, SolidWorks, Creo
Search, index, cite
- Reads through standard interfaces
- Understands requests and documents
- Geometry index for part shapes
- Every hit with its source
- Writes nothing back
Read-only by design. We read from your systems, we do not write into them and we do not replace them. Engineers keep working in the tools they know; the AI adds the answers they could not find before.
inteliScience is a Siemens NX, NXOpen and Teamcenter specialist, and this service is deliberately CAD-neutral: it works with Siemens NX, CATIA, SolidWorks, Creo or a mix of them, and with commercial or self-built PDM systems. This hub explains what AI for engineering data can do, what decides whether it works in your company, and how a project starts. The two services below go into detail.
What engineering data AI means in practice
A new request arrives, and somewhere in the company there is a completed project that already solved most of it: its specification on a file server, its parts and assemblies in the PDM, its costs in the ERP, its notes in a wiki. Whether anyone finds it usually depends on who still remembers it. Engineering data AI makes that knowledge searchable across all of these systems at once.
- Reads the request: analyses an incoming RFQ package (requirements specification, technical PDFs, drawings and further requirements) and recognises the technical requirements and features that matter.
- Finds similar past projects: searches completed projects and shows which ones are technically similar and why each was rated similar.
- Cites its sources: shows which documents and information exist for each hit, with the direct source passages.
- Links commercial data: connects hits to PDM records and to ERP article, cost and procurement information.
- Finds parts by shape: detects similar and duplicate parts from their geometry, regardless of how they were named or numbered.
A PDM search already finds a part when you know its number or name. It cannot read a new specification and tell you which completed project solved a comparable problem, and it cannot find the part that has the same shape under a different name. Engineering data AI adds those two questions on top of the systems you already have.
Two services built on your engineering data
Each service has its own page with the full workflow, the data it needs and its limits. Both use the same read-only access to your systems, so you can start with one and add the other later.
- Requirements and features recognised in the RFQ package
- Documents and source passages for every hit
- Linked PDM records and ERP cost and procurement data
- One-time index built for your part library
- CAD-neutral: Siemens NX, CATIA, SolidWorks, Creo
- Conversion on your infrastructure, CAD data stays in-house
Geometry search tools on the market, such as CADENAS or Siemens Geolus, compare shapes. They do not analyse a specification, find similar projects, cite source passages or link hits to the costs and quotations of past projects; that needs a solution tailored to your data. We can take overall responsibility for both, from the geometry index to the project search.
Why the data, not the AI model, decides
The technology for reading PDM, ERP, wiki and file server data is well understood. The real uncertainty in every project is data quality: does a project or order number run consistently through all of these systems? That single question decides how easily a hit in one system leads to the parts, costs and documents in the others.
One project or order number, four systems
The number runs through every system
The same project or order number appears in the PDM, the ERP, the wiki and the folder structure. A hit in one place leads straight to everything else.
- PDMnumber present
- ERPnumber present
- Wikinumber present
- File servernumber present
The number breaks somewhere
It is missing in one system, kept in a different format or replaced by a name. Still solvable, but the linking becomes part of the project scope.
- PDMnumber present
- ERPnumber present
- Wikimissing or different
- File servernumber present
This is why we look at the data before we talk about effort, and why it is the central topic of the Quick Scan with an on-site workshop we recommend before anything is built. Effort depends mainly on the quality and linkability of what already exists, not on which AI model is used.
Read-only access to the systems you already run
No migration, no new master system, no write access. Each source is read through the interface it already offers:
| Source | How it is read | What it contributes |
|---|---|---|
| PDM systemcommercial, or self-built on a SQL database | Read access only; a PDM on a SQL database can be read directly | Part and assembly numbers, names, product structure and project assignment |
| ERPfor example SAP S/4HANA, SAP Business One or proAlpha | Read-only through the vendor’s standard interfaces | Article, cost and procurement information for each hit |
| Wikifor example Confluence or SharePoint | Standard connectors | Project notes and documented know-how |
| File serversnetwork shares and project folders | Standard connectors | Specifications, technical PDFs, drawings and further project documents |
| CAD geometrySiemens NX, CATIA, SolidWorks, Creo | Converted once into a neutral format, on your infrastructure and with your CAD licences | The geometry index for similar and duplicate parts. Not needed for a project search on PDM metadata |
For the geometry index, the conversion typically runs on a remote workstation or virtual desktop that you provide. As a service provider we hold no CAD licences, which is normal in such projects and means your CAD data never leaves the company. Where the AI itself runs, and who may see which data, is settled with your IT before a proof of concept starts. Self-hosted AI on confidential documents is not new territory for us; our private AI knowledge assistant case study describes a RAG assistant that runs inside a client’s own infrastructure.
How an engineering data AI project starts
Small and on your own data first. A bounded proof of concept shows more than any demo on someone else’s projects.
-
Step 1 · one week
Quick Scan with on-site workshop
Two workshop days at your site, five days in total. We look at your current situation together: which systems hold what, and whether a project or order number runs consistently through them.
-
Step 2 · bounded
NX Sprint
One product family and a handful of completed projects. Real requests run against your own data, with hits, reasons and sources you can check.
-
Step 3 · your decision
Rollout, or not
The effort for more product families and more sources is estimated from your data, not from assumptions.
Entry point · bounded scope
NX Sprint
A bounded proof of concept on one product family and a handful of completed projects, run on your own data: the step between a demo and a rollout.
- Read-only access to the systems that matter for the scope
- Similar projects, similar parts, or both, on your own data
- Every hit with its reason and source passages
- A clear view of the data quality work a rollout would need
Read-only integration · CAD-neutral · CAD data stays in your company
What helps from your side
Inputs for the sprint
- One product family and a handful of completed projects
- Read access to PDM, ERP, wiki or file server, as far as the scope needs
- For geometry: a remote workstation or virtual desktop with your CAD licences
- An engineer who can judge whether a hit is really similar
Guides on engineering data and AI
Seven articles that go deeper into single questions, from reading a requirements specification to preparing CAD data for an index.
Projects, specifications and costs
Parts and geometry
Related services
Engineering data AI shares its foundation with our knowledge assistants and quotation services: answers from your own data, with sources, under your control.
Frequently asked questions about engineering data AI
What is engineering data AI?
Can AI search PDM, ERP and wiki data without replacing those systems?
How does AI search for engineering documents show where an answer comes from?
Does AI need CAD files to find similar engineering projects?
What is the biggest risk in an AI project on engineering data?
Does CAD data have to leave the company for AI geometry search?
How much effort does an AI project on PDM and ERP data take?
What is the difference between CAD geometry search tools and AI search over engineering documents?
Name one product family and a few projects you have completed. We will tell you what your data can answer.
In a Quick Scan with an on-site workshop we look at your systems and at whether a project or order number runs through them, and we tell you honestly whether an NX Sprint makes sense.
Talk to an EngineerRead-only integration · CAD-neutral · CAD data stays in your company · Siemens NX, NXOpen and Teamcenter specialists

