RouteMe.AI — Humans review the paths before the AI ships them
The admin platform behind an AI wayfinding product — organisations, locations, stations and route networks, built for high-density data.
[ The problem ]
An AI wayfinding product needs an operations layer its own team can trust. Generated routes are fast but not automatically correct, and a wrong path in a hospital or an airport is not a cosmetic failure. The admin platform had to hold high-density data — organisations, locations, stations, route networks — for three different roles without any of them drowning in it.
[ Goals ]
Review before release — no generated path reaches a user unapproved. Density without noise — organisations, locations, stations and routes in one hierarchy. Role clarity — admin, operator and client each see a different truth from the same data.
Approach
The platform was structured around the real hierarchy — organisation, location, station, route — so that navigation matched how the business thinks rather than how the database is shaped. Each role was designed as a different truth from the same data: what an admin, an operator and a client each need to see. The review surface came first in priority even though it sits second in the flow, because it is the product decision the rest depends on: a generated path, its confidence, what it connects, and approve or reject as the primary actions. Analytics followed, designed to be read quickly rather than explored slowly.
[ Scope and constraints ]
[ The work ]
The surfaces.
Organisations and locations
Give one admin a workable hierarchy across many client organisations, sites and stations.
Structured the platform around the real hierarchy — organisation, location, station, route — so navigation matches how the business actually thinks.
An operator can move from a client to a single station without losing their place.
Route review
Put a human between the AI and the person following the directions.
Designed a review surface that shows the generated path, its confidence and what it connects, with approve and reject as the primary actions.
No generated path reaches the navigation app without approval.
Analytics
Show which routes are used, which fail, and where people give up.
Built a dense analytics view for QR scans, station performance and route completion, designed to be read quickly rather than explored slowly.
Route quality becomes measurable rather than assumed.
[ The system underneath ]
[ Results ]
3 | Roles, one data model 100% | Generated paths reviewed before release 4 | Levels of hierarchy in one navigation
No generated path reaches the navigation app without a human approving it, and the approval is fast enough that nobody routes around it. Operators can move from a client to a single station without losing their place, and route quality is measurable rather than assumed. The review gate is the product decision that mattered. An AI wayfinding system that publishes unreviewed paths is a liability in a hospital or an airport; one that routes every path past a person first is an asset.
What I carry forward
Designing the human gate first taught me that the most important screen in an AI product is often the one where a person says no. Make that fast and obvious and the rest of the system earns trust by default.