Inside GEHA's Smart Manufacturing Stack: AI Drawing Review, C# Automation and Data-Center Fan Production Run by One Engineer

Smaller teams do not automatically mean slower engineering. In GEHA's drawing office in Nieuw-Amsterdam, the Netherlands, a verification step that used to consume a weekend is now a standard gate on every large order — and one person operates it. The fan maker has been at work since 1957, with housings reaching up to 7.3 meters in diameter; for most of that period the buyers were refineries, chemical plants and the heat-transfer suppliers that serve them. Today the newest customer category in the order book is the data center.

For a global manufacturer of datacenter cooling systems serving the North American market, GEHA is producing fans of more than 2.1 meters (7 ft) in diameter in high volume, and a single datacenter campus can require several hundred of them. Jan Bos, the company's manager of engineering and R&D, states the positioning without hedging: “We make the biggest and most silent fans for data centers,” he says, and adds, “That’s our specialty.”

What makes the case legible for other plants is the staffing behind it. The engineering department once had four people; when the oil market turned in 2017 it shrank to two. One engineer now carries the workload, running Designcenter Solid Edge plus a growing kit of AI tools. “I’m on my own and running the whole show without problems,” Bos says, and adds, “And I’m enjoying my work.”

GEHA supplies industrial air-cooling solutions. (Image: GEHA.)

Credibility first: a community founder who never became a coder

The AI layer sits on decades of CAD use. In 1997, on version 3.5, Bos first saw Solid Edge: he had been working in 2D CAD and was unimpressed, until a colleague’s son modeled a part for him in an afternoon and placed views, a detail and a cross-section onto a drawing in seconds. That introduction to 3D CAD settled the argument. In 2003 he brought Solid Edge to GEHA to move the company’s knowledge “out of the brains, out of the paper, into the model.”

Since then he has helped found the Benelux user group, become one of Siemens’ Community Catalysts, and presented at Realize LIVE on subjects ranging from Visual Basic automation to a refinery retrofit GEHA engineered from a point cloud. This year in Detroit, his talk was titled “Practical AI as an Everyday Solid Edge User.” His introductory slide reads: “Not a programmer. Just an engineer who got curious.”

Jan Bos, lead engineer and designer at GEHA, presents his AI design hacks to the community at Realize LIVE Detroit 2026. (Image: GEHA.)

Gate one: an AI review step that replaced a manual ritual

Bos’s original quality step was a private one. On large projects he printed the drawings and checked them at home on Sunday mornings with a red pen, a highlighter and a cup of coffee; a 150-drawing project took about two to three hours. “It only goes wrong when it’s a big project,” he says, and adds, “So, if it’s a big project with a lot of parts, you have to check it very well.”

The rewrite came in spring 2025. A full set was exported to PDF and loaded into Google AI Studio running Gemini 2.5 Pro, with the model told to act as the kind of old-fashioned drawing checker engineering departments used to employ, paying close attention to the bill of materials. Bos’s own shorthand for the before-and-after is a Sunday morning versus two minutes.

The findings were specific rather than generic. The model detected uneven hole spacing on a perforated plate and asked whether it was intentional; it was not. It flagged a transport wheel listed in the bill of materials with no diameter and suggested 150 mm, a value pulled from the detail drawing. “The AI did what a real drawing checker would do,” Bos says.

Typical fan in the GEHA product line: machine-produced fan rings are up to a diameter of 7.30 meters, with components such as drives, bearing blocks, louvre dampers and plenums. (Image: GEHA.)

That pass is now standard on every major GEHA project, and it has been promoted from a one-off prompt to a reusable skill. Bos taught the checker who does what in the factory — fan construction, assembly, work preparation and laser cutting are all represented — so feedback arrives framed around the people responsible for it.

Bos says, “[AI] tells me, ‘Jan, you made a heavy construction, but there is no hoisting lug on this one, and your colleague Eric will not be very pleased,'” and adds, “It holds up a mirror.”

The scope extends past paper. Given a PDF or a STEP file, the model will flag a hole placed too close to a bend, where forming will distort it — design-for-manufacturability feedback that once depended on a shop-floor veteran catching it. Bos has applied the same approach to checking GEHA’s technical construction files against updated machine-safety regulations. Does he trust it? “Yes,” he says, and calls the results “very solid.” Part of that trust is scope: the model checks his drawings against his standards rather than roaming the open internet. The other part is that the AI doesn’t make the call. As one of his Detroit slides puts it, “AI amplifies expertise. It doesn’t replace it. You still need engineering judgement.”

Gate two: turning macro debt into maintainable automation

The second front is code. Like many long-time users, Bos had accumulated VB6 and VB.NET macros that “sort of worked,” built by adapting examples. Using the open-source coding tool OpenCode with a paid Claude model, he converted an old VB6 utility called PM-Tabelle into Fit Table, a C# macro that connects directly to Designcenter Solid Edge without opening Visual Studio. Every GEHA macro is now C#. When the AI needs to understand Designcenter Solid Edge’s programming interface, it reads the programming documentation available online; as a starting point, Bos can hand it one of the roughly two dozen example programs he has collected over the years.

He probed the ceiling with a property add-in that launches on his second monitor, something he had wanted for years. After a few failed attempts, the AI asked whether he had other add-ins it could study, found one in a folder and produced a working version; Bos refined the layout by marking up screenshots with red arrows in Microsoft Paint. Total concept time was about an hour and a half — an experiment that once would have cost five or six days of a contract programmer’s time. He ultimately preferred his own macro and shelved the add-in.

The same method cleared a bottleneck downstream of the model. A colleague used to spend half an hour on each of roughly 20 orders a week splitting Designcenter Solid Edge bill-of-materials reports into lists for purchasing, laser cutting and other departments. Bos gave Claude Code an emptied-out report as a template, and it built a Python tool that does the job with a right-click.

Some newer work skips CAD entirely. With Claude connected to SketchUp, Bos can photograph a dimensioned hand sketch of a bracket and get back a STEP file ready to import into Solid Edge or send to a 3D printer. In the other direction, he has handed the model a STEP file of a GEHA bearing block and received a bill of materials, plus a conversation about bearing selection and where the part could be made cheaper.

Bos went from whiteboard to STEP 3D model without opening a CAD tool. (Image: Jan Bos | GEHA.)

Architecture: the model is a layer, the CAD base is not

For all of this, Bos is clear about what holds it together: AI is the layer on top, and Designcenter Solid Edge is the base. “Solid Edge is a very, very stable building block,” he says, and adds, “Everything we communicate to suppliers, it’s always spot on.”

What he wants now is for the software to absorb the work he has been bolting on. He says, “The nice thing about being an engineer is the creating part, solving the puzzle,” and adds, “We don’t like making a drawing, checking a drawing, printing drawings, doing conversions. All that stuff should be handled by Solid Edge.” His Detroit wish-list starts with native batch drawing management:

Batch drawing-management capability	Where it sits today
Gather every drawing	Manual step outside the CAD tool
Verify it is current	Manual step outside the CAD tool
Update hundreds of drawings at once	The core ask; not available natively
Report what changed	Left to the engineer’s own review

That would build toward AI that drafts and checks drawings alongside the engineer, ideally on a local model so data never leaves the building. Siemens is working toward AI-driven drawing creation and checking, and Bos sees Designcenter Solid Edge 2026 already heading that way. He would just like it faster. “We are getting spoiled,” he says. “Why not?”

Jan Bos, third from the left, joins fellow Community Catalysts and Siemens staff at Realize LIVE Americas 2026. (Image: GEHA.)

Implementation path: how the stack was assembled

Encode a ritual you already run. The first automated check simply reproduced Bos’s own red-pen review, so the rules were defined before any prompt was written.
Export a finished package. In spring 2025 the complete set went to PDF and was loaded into Google AI Studio running Gemini 2.5 Pro.
Assign the model a narrow role. Behave as an old-fashioned drawing checker; watch the bill of materials closely.
Triage what comes back. Uneven hole spacing that was not intentional; a missing transport-wheel diameter answered with 150 mm from the detail drawing; manufacturability risks such as a hole placed too close to a bend.
Promote the prompt to a reusable skill and add the role map. Construction, assembly, work preparation and laser cutting now shape how findings are phrased.
Extend the same gate to compliance. Technical construction files are checked against updated machine-safety regulations the same way.
Only then automate tooling. Inventory legacy VB6 and VB.NET macros, convert them one at a time to C#, and keep the CAD model as the system of record.

Selection checklist before an AI gate goes live

Does the model check against your standards and your files rather than open-internet content?
Does the final decision stay with a person — “AI amplifies expertise. It doesn’t replace it. You still need engineering judgement.”?
Is the CAD base stable enough that supplier-facing output stays spot on?
Can the tool read both PDF and STEP, so drawings and models fall under one rule set?
Does it know the shop’s roles well enough to route a finding to the colleague who will act on it?
Is there a credible path to a local model for projects where data must not leave the building?

Common pitfalls

Automating before documenting. A library of VB6 and VB.NET macros that only “sort of worked” is debt, not an interface.
Letting the tool decide. The review earns trust precisely because it does not make the call.
Open-ended scope. Reliability came from restricting the model to GEHA’s own standards.
Ignoring the handoff after the model. The bill-of-materials split consumed half an hour per order for roughly 20 orders a week long after design work was finished.
Reading headcount as capability. Going from four people to two, and then to a single engineer, did not remove the need to check large packages rigorously.

FAQ

Does the engineer have to know how to program?

No. The opening slide says it directly: “Not a programmer. Just an engineer who got curious.” The AI reads the published programming documentation, and Bos can point it at one of the roughly two dozen example programs he has kept over the years.

Is AI review dependable enough for production drawings?

Bos answers “Yes” and rates the output “very solid,” on two conditions: the check runs inside his own standards and files, and the engineering judgement remains his.

What still has to come from the CAD platform?

The model and the drawing set remain the communication layer with suppliers, which is why gathering every drawing, verifying currency, updating hundreds of drawings and reporting what changed top his wish-list.

How quickly did the pilot become routine?

What began in spring 2025 is now standard on every major project, replacing hours of manual review on a 150-drawing package.

Action advice for manufacturing teams

Pick the one review your best engineer performs manually and write it down as rules before touching a prompt.
Convert legacy automation incrementally: one “sort of worked” macro per cycle, targeting C# against the documented interface.
Feed the checker your role map so findings reach the assembler, the laser-cutter or the work-preparation planner in usable language.
Watch the steps that sit outside CAD — report splits, conversions, file gathering — because those are both the quickest wins and the gaps you should raise with your vendor.
Protect time for tooling. Bos keeps Friday afternoons free for “sharpening the tools.”

The through-line at GEHA is not a tool list but a decision pattern: keep the CAD model as the base, keep judgement with the engineer, and let AI absorb only the checking, conversion and clerical work that was never the interesting part of the job. That framing also explains why the work still feels new to Bos, taking him back to that afternoon in 1997, when a cross-section appeared from a 3D model in seconds. “It’s that magic feeling I had in ’97,” he says, and adds, “It’s a multiple of that magic at the moment.”