AI adoption
that survives
contact with
reality.
Designing Industrial AI & Control
Systems for Better Operations.
AI is the next instrument of optimization — the same goal I've pursued for eighteen years, with a new tool. The discipline that keeps a plant running safely is exactly what AI adoption needs.
Demos dazzle. Production humbles.
The gap between “look what it can do” and “it quietly does this every day” is where most AI pilots die. Closing it is a discipline, not a demo.
Industrial or not — same questions
Every organization adopting AI faces the problems I've spent a career solving on the plant floor.
The value is in the seam
Real operations and modern AI rarely share a person. I work where the two meet — and have to hold.
Agentic AI, meet the plant floor.
Most industrial operations don't need to rip anything out to adopt AI. The data is already there, in the PLCs, the SCADA tag database, and the historian, generated every second of every shift. What's missing is the path from that operational layer to a place where modern AI can actually use it, and the discipline to do it without putting the control network at risk.
That path is an OT-to-cloud pipeline. Equipment exposes its values over OPC UA, a lightweight service streams them into a queryable store, and a retrieval layer grounds an AI agent in that real history instead of guesswork. Done correctly, read paths and write paths stay separate, safety logic is never touched, and the data leaving the plant is the data, not a lossy summary.
On top of that foundation sits the part people mean when they say agentic AI: systems that don't just answer questions but plan a task, call the right tools over standard protocols, and carry work through with a human in control of anything that matters. Industrial RAG lets an engineer ask, in plain language, why line 3 tripped last night and get an answer grounded in the actual trends. Predictive maintenance moves the operation from reacting to failures toward seeing their signature early.
The hard part was never the model. It's the integration with legacy SCADA and PLC systems, the validation, and the safe-failure design that 18 years in safety-critical control systems teaches you to insist on. That is the seam this work lives in: bringing Agentic AI and OT-to-cloud architecture to operations that can't afford to fail, for clients across the US and the Nordics.
Eighteen years of automation.
Two years of modern AI.
Automation that can't fail
Control systems on offshore platforms, safety systems in pharma plants, machining lines at one of the world's largest truck manufacturers. PLCs, SCADA, commissioning from first wire to final sign-off.
Agentic systems & RAG
Two years building, not spectating: agentic systems, retrieval-augmented generation over messy real-world data, structured LLM integration, model evaluation, fine-tuning, and local inference.
Trustworthy by design
In safety-critical work, a system that breaks in production isn't a disappointment — it's a hazard. You learn how to make automation survive the messy, high-stakes conditions of the actual floor.
Where operations meets AI
Most people have one half — real operations or modern AI. The value is in the seam: getting these systems to do useful work reliably, in the place where they meet reality and have to hold.
Projects
Systems where operations meet AI — an industrial RAG and OT-to-cloud pipeline, and an open-source operating system for an AI-run company.
Articles
Notes from the seam between operations and AI — on moving plant data to the cloud, optimization versus automation, and AI as the next instrument.
Got a real problem?
Get in touch.
No pitch, no package. If you're trying to adopt AI for real — in operations or anywhere else — tell me the actual situation and I'll tell you honestly how I'd approach it.
EMAIL USMAN