- Entry date
- 1 August 2025
- Category
- Tooling
- Access
- 🔓 Public
A functional-but-plain passenger display became designer-quality with AI in the loop — on a screen where legibility under pressure is a safety constraint, not a preference.
- React/TypeScript + Vite
- frontend stack
- AI designer
- assisted theme customisation feature, shipped
Honest evaluation
The improvement is real and delivered, but reported as a qualitative shift ("functional-but-plain to designer-quality"), not a measured figure — stated as such rather than inflated into a number.
What would prove or disprove it further
What would prove or disprove it further: a structured before/after design review (task-completion time, error rate, or a panel rating) against the same screens would convert "designer-quality" from a qualitative judgement into a measured one, and either strengthen or puncture the claim with a number instead of a description.
The evidence — full reasoning behind the verdict
Verdict: proven. The claim was that AI-assisted design tooling could meaningfully improve a safety-relevant passenger display's visual and UX quality without taking judgement away from a human designer — and a shipped, working, delivered system carrying an AI-assisted "designer" feature and a drag-and-drop widget layer is real evidence for that, not a demo reel.
The evidence is qualitative rather than a number, and the report says so directly: the result is described as moving the PIDS "from functional-but-plain to designer-quality," not as a measured percentage improvement. That's an honest way to report a genuinely subjective outcome — legibility and visual coherence aren't naturally single numbers — but it does mean this verdict rests on delivered, working software and a stated qualitative judgement, not a hard metric. The boundary the report holds carefully — AI assisted the designer's decisions, it did not set the legibility/contrast/glanceability constraints itself — is exactly the claim being tested, and it holds.
A passenger information display that works is one thing; a passenger information display that's legible, coherent, and actually well-designed under real operating conditions is a different, harder bar. This project put AI design assistance directly in front of that bar.
What it was
An advanced Passenger Information Display System (PIDS) built for a global transit passenger-information programme — the operator is withheld by name per our policy of never naming clients in safety-critical/transit work. The system included AI-powered design assistance and a drag-and-drop widget system for building display layouts.
What we built
Shipped & measuredA working React/TypeScript frontend on Vite, with an LLM API driving an "AI designer" feature: theme customisation and layout suggestions generated with AI assistance, on top of a drag-and-drop widget system for assembling what actually appears on a passenger screen. The system shipped and worked.
What we learned — including the honest negative
The real result was qualitative, not a number: this took a PIDS from functional-but-plain to designer-quality, with AI meaningfully assisting visual and UX decisions rather than just generating boilerplate code. That's a genuinely different kind of AI-assistance claim than "it wrote the CRUD layer" — here it was helping shape what a passenger actually sees and reads at a glance.
The honest edge is exactly that boundary: design assistance, not design autonomy, on a display surface where the cost of getting legibility wrong is real.
Where it went / status
Working and delivered, sitting alongside the operations-control side of the same domain (see the rail PA/PIDS platform) as one of two production passenger-information systems built the same year — one running the control plane, this one making the screens themselves better designed.
What is still open — kept visible
The honest edges, next to the wins. This is what turns 🔬 into 🟢 — honestly.
- The transit authority this was built for is withheld by name, per our no-client-names policy — described here only as "a global transit passenger-information programme."
- AI-powered design assistance augmented a human designer's decisions; it did not replace design judgement, particularly for the safety-relevant legibility constraints a PIDS has to meet.
Where this connects
Sources
- vocabotics project audit — Transit PIDS AI design system (client withheld), Aug 2025vocabotics internal project history · as of August 2025