Sixteen years on both sides of the dashboard.
I spent a decade running supply chains before I started analysing them. Now I build the tooling that lets other people stop asking me for numbers.
LLM agents that let non-technical teams pull warehouse data from a plain-language question, and a Slack bot answering routine stock and purchasing questions with no analyst in the loop. Next: the functions still done by hand — master data and replenishment.
Designed and launched a supplier-facing analytics portal that turned an internal reporting asset into something the business sells. Most reporting is a cost line. Occasionally it doesn't have to be.
A data literacy programme run with department leadership — training, licences, ongoing support — so teams pursue their own questions as far as their role allows, instead of handing them to an analyst and forgetting about them.
My first job was analysing warehouse data at a steel plant. Then I spent ten years actually running supply chains — stores, distribution centres, S&OP, third-party logistics, and a sixty-store ERP migration where I was the client rather than the consultant.
That decade is why the tooling I build gets adopted. Any competent engineer can connect a language model to a data warehouse; that part is a weekend. The hard part is knowing which of the forty columns in an inventory table a planner actually trusts, and why they quietly stopped trusting the other thirty-nine. That knowledge is not in the documentation.
The analysis I am proudest of usually contradicts the request that triggered it. That a reorder quantity ignored the attributes actually driving demand. That a reported data failure was really a categorisation problem belonging to another team. That a feature someone asked me to build should not be built at all.
Industrial engineer by training.
Order picking has some of the highest staff turnover in warehousing, and gamification is the standard prescription. The difficulty is that nobody can separate a well-designed game from a badly-designed one until it is already on the floor and a shift has been spent finding out.
My thesis is an attempt to move that test earlier — to score a scenario before it is built, and measure it before it is deployed.
A full picking floor rebuilt in Unreal Engine. Candidate scenarios are scored against the eight Octalysis drives, built onto the same simulated floor, and measured on travel time, distance, waiting time and picking errors — so a design fails in the simulator instead of on a real shift.
unrealwarehouse.hejrani.comWhy AI in operations is a knowledge problem wearing a technology costume — and why the functions with the most to gain are the ones nobody wants to spend a decade inside.