Hector Rail · European Rail Freight

Hector Rail

Proving AI-first delivery could be trusted, not just fast.

Hector Rail came in as an inbound revamp lead from an existing customer. Rather than run it on the standard timeline, I proposed an AI-first delivery model built on HMHT, compressing a multi-month engagement into a matter of weeks, requirements through frontend.

The problem

One business, fragmented across country sites.

Hector Rail operates as one business, but its web presence didn't read that way. Each country site had evolved independently, different visual language, structure, and tone, so a visitor moving between country sites had no sense they were looking at the same company.

The brief was to make every country site speak one consistent design language, with a language switcher handling localization as a separate concern from that visual consistency problem.

Approach

Three decisions that made speed defensible.

The brief wasn't just "go faster." It was going faster without leadership having to take that on faith.

Decision 01 · Load-bearing

Proposed an AI-first delivery model instead of the standard phased process.

Rather than accept the default multi-month timeline, I made the case that requirements, design, and frontend delivery could run as one compressed, AI-assisted track, restructuring how the early stages worked, not just working faster within the existing structure.

Decision 02

Used HMHT across the full engagement, not selected features.

Unlike iGOT, where HMHT was scoped to specific decisions, Hector Rail ran it end to end, simulating user reasoning across requirements, analysis, and design where there wasn't time for a traditional research cycle, validated against client stakeholders as we went.

Decision 03

Used Google's AI Studio for range, then engineered a consistency layer on top.

AI Studio gave design range that would have been hard to match manually, but had no memory for consistency across generations. The fix was a dedicated prompt keyed to the established design language, locking new outputs to match what already existed.

Outcome

What changed, and what it proved.

Weeks, not months

Requirements through frontend delivery compressed from a standard multi-month timeline.

Repeatable model

Established leadership confidence that AI-first delivery is a model that can be used again.

One language, every market

All country sites unified under one design language, localized through a dropdown.

Reflection

Speed came from a deliberate trade, simulated reasoning carried the early decisions instead of a full research cycle, validated against client stakeholders as we went. That's exactly the trade AI-first delivery is supposed to make, and it held up. The next phase is layering real end-user validation on top of a model that's already proven it can move this fast.