Shipped predictive software for industrial location tracking — at Cambridge's RTLS company.

RTLS shows you where things are. Customers wanted to know where they'd be.

Ubisense's core product locates physical objects — tools, vehicles, components — to centimetre precision in real-world industrial environments. Automotive plants, smart factories, defence. The data is rich. The interpretation, historically, was reactive: customers reacted to where things were, not where they were heading.

The brief I picked up sat one layer up: predictive tooling that turned live tag streams into operational signal — anticipate bottlenecks, surface risk, give floor managers something to act on before the problem became visible to the eye.

Ship a useful version of the prediction. Then ship the interface.

Two parallel tracks. The predictive layer needed to work against real customer tag streams — not synthetic benchmarks. The interface needed to be usable by people who don't want a model output, they want an answer.

Production code in. Invitation back.

Industrial software is unglamorous from the outside and ferociously hard up close. The product has to work in real environments with real consequences. The internship taught me how serious people build serious systems — and how much of that is judgement, not code.

The interface is half the product. Sometimes more.

The hardest part of predictive tooling isn't the prediction. It's the moment a busy human sees the output and either acts on it or ignores it. That decision is design, not maths. The Ubisense experience cemented that for me — and it informs the way I build internal tools today, where one good glance beats a perfect dashboard.