Manufacturing process design and MES development, working with AI
I'm Yuval Goldenberg, a process and MES architect with more than 15 years in plants. I design processes that cross the organization, from purchasing to finance, in SAP and on the shop floor, and working closely with AI I build the systems that run them: on open source, inside your plant, in about a quarter of the time.
- Purchasing
- Planning
- Production
- Automated warehouse
- Quality
- Maintenance
- Finance
Among Grovv's clients
Process first. Then code.
AI writes code fast. It doesn't know how a purchase order becomes stock, how a production order settles in finance, or what the automated warehouse does when an inspection lot is blocked. I do: I've been designing these processes for more than 15 years, in SAP and on the shop floor.
A purchase order raised from MRP, not from a spreadsheet, is the difference between planned stock and accidental stock.
With incoming inspection, material goes into quality-inspection stock and isn't available to production until the usage decision.
MRP plans from the BOM and the routing. A part with wrong master data is planned wrong, every night.
The line reports to SAP by itself, and yield is computed from deltas, because the PLC counter resets at shift change.
A line stop opens a maintenance notification on the right equipment, with the downtime, not an email.
A result from the inspection station is recorded on the right lot, and the decision on it changes stock immediately.
A task sent to the crane that fails has to come back to SAP, or system stock and rack stock drift apart.
Every confirmation and goods movement ends up in finance. A double posting on the floor is a cost variance.
- PurchasingPurchase order
A purchase order raised from MRP, not from a spreadsheet, is the difference between planned stock and accidental stock.
- Receiving & inspectionGoods receipt · inspection lot
With incoming inspection, material goes into quality-inspection stock and isn't available to production until the usage decision.
- PlanningMRP · production order
MRP plans from the BOM and the routing. A part with wrong master data is planned wrong, every night.
- ProductionConfirmation · goods issue
The line reports to SAP by itself, and yield is computed from deltas, because the PLC counter resets at shift change.
- MaintenanceMaintenance notification
A line stop opens a maintenance notification on the right equipment, with the downtime, not an email.
- QualityIn-process inspection · usage decision
A result from the inspection station is recorded on the right lot, and the decision on it changes stock immediately.
- Automated warehouseWarehouse task
A task sent to the crane that fails has to come back to SAP, or system stock and rack stock drift apart.
- FinanceOrder settlement · variances
Every confirmation and goods movement ends up in finance. A double posting on the floor is a cost variance.
Master data: part · BOM · routing
In SAP ERP: production and planning · inventory and master data · automated warehouse · quality · maintenance · purchasing · finance
And when the product flows: from the tank to the pallet
In process manufacturing you don't count pieces, you follow material: which tank is feeding the line now, which one is being cleaned, and what is ready for the next fill. And it isn't only line control. The system reports to SAP on its own: yield, material consumption, goods movements 261 and 101, and order status changes.
- Tank 1Filling5,160 L · 43%
- Tank 2CIP cleaning0 L · 0%
- Tank 3Feeding line 85,160 L · 43%
- Tank 4Ready10,320 L · 86%
- Process order
- 1118002
- Bottles
- 1,389
- Pallet
- 5 / 8
- 261Material consumption for the order10,320 L · Tank 21118002
- CONFYield confirmation1,382Bottles1118002
- 261Material consumption for the order10,320 L · Tank 11118002
- STATOrder status changeREL → PCNF1118002
There is now a third way to build a shop-floor system
Until recently a plant had two options: buy a large commercial package and fit the plant to it, or fund a long custom project with a whole team. AI changed the arithmetic. Today you can build exactly what the process needs, on open source, in about a quarter of the time.
A commercial package
- Licenses and maintenance fees every year
- The plant adapts to the software
- Every change goes through the vendor
Custom development, the old way
- A whole team, for months
- Testing pushed to the end
- Knowledge leaves with the people
A process architect with AI
- Only what the process needs
- Tests for every interface from day one
- Open-source code that stays with you
Illustration: the same scope, delivered by a process architect working with AI in about a quarter of the time of a conventional project, with a larger share of it spent on testing.
AI multiplies whoever uses it
Someone building a system for the first time will write a lot of code with AI, fast. A process person who has managed and led implementations on about eight production floors knows what to build, what to leave out, and where it will break at three in the morning.
| A developer with AI | A process architect with AI |
|---|---|
| Writes code fast | Knows which code not to write |
| Sees the screen they were asked for | Sees the whole process, from purchasing to finance |
| Learns the floor on your line | Knows it from nights on Kepware and hours on BarTender |
| Closes features | Closes processes, with the people who run them |
I'm Yuval Goldenberg, and processes are my trade. For more than 15 years I've designed processes that cross the whole organization, from purchasing through planning and production to finance, and I know SAP ERP in depth: production and planning, inventory and master data, the automated warehouse, quality and maintenance. I've managed and led shop-floor system implementations on about eight production floors, in both process and discrete manufacturing. I've spent nights on Kepware and hours on BarTender, and designed and built dozens of machine interfaces. For the past two years I've worked with AI every day, all day. I see quickly what matters and what doesn't, and I stay on a job until it works.
You stay in control
AI works for me while I build. The system, the code and the data stay yours.
- 01
Full control of the processes
The system is built around your processes, and changes when they do.
- 02
Full control of the code
The code I write is open, and handed over with documentation and tests to your own repository. You don't depend on a vendor, or on me.
- 03
Full control of the pace
You see progress as it happens, and decide what goes live first.
- 04
AI in development, not in the system
AI helps me write and test the code. The system you receive runs without depending on it.
- 05
Runs on site, no cloud
Installed on your servers, inside your network, with no internet dependency.
- 06
Production data stays in
While building I work on code and sample data. Real production data is never sent to AI services.
What I've built in the last two years
Built with AI at my side, every day.

Mekorot's water balance
Mekorot's water balance, redeveloped in SAP's CDS layer.
- SAP CDS
A local UMH deployment
United Manufacturing Hub, an open-source Unified Namespace platform, installed on the plant's own network.
- UMH
- MQTT
- Unified Namespace
Also built with AI, outside the plant
- YG_BOT · automated trading systemThree strategies running in parallel on live market data, a live dashboard over WebSocket with performance metrics, and an AI advisor that ranks the strategies and makes a call. Runs in simulation.
- WebSocket
- Real-time
- AI advisor
- campro.co.il · outdoor gear storeA WooCommerce store for outdoor and camping gear, with a theme I built with AI, a Caspit connection that publishes products, and branded Hebrew email templates.
- WooCommerce
- Caspit
- Cloudflare
Before that: MES at Tempo, Nilit, PCB and more. See the clients
And also: advising management through implementations
Alongside design and development, I also advise management through implementations, even when someone else builds the system: on the steering committee, as an advisor to the CEO and as a mentor to the project manager. With the experience of about eight production floors where I managed and led the implementation.
- 01On the steering committee
- 02Advisor to the CEO and management
- 03Mentor to the project manager
What plant owners ask me
You're one person. What happens if you're not available?
Can code written with AI run a production line?
Does our data leave the plant?
Why open source rather than a commercial package?
Will it connect to our SAP?
We run SAP ME or MII. What now?
We've already chosen a vendor. Can you still help?
How do we start?
Tell me about one process that takes too long
One conversation, at no cost and without slides. I'll tell you what I'd build, what I'd leave out, and in what order.






