AETHER EV · Process

The full process

The research, personas and flows behind AETHER EV. The case study covers the decisions; this is the work underneath them.

02 — Discovery & Research

Where the friction actually lives

The competitive review made clear that incumbents optimise the wrong moment. They invest heavily in the remote-control novelty — honk, flash, precondition — and leave the daily decisions unsupported.

The daily question is small
Not 'what is my range' but 'can I do today's plan without thinking about it'. That is a different, harder question.
Trust compounds or collapses
One confidently wrong range estimate poisons every subsequent one. Accuracy matters less than legibility.
Connectivity is unreliable
Underground car parks and rural charge points are exactly where the app is needed and exactly where the network is not.
Segments diverge sharply
The commuter, the road-tripper, the fleet manager and the enthusiast want genuinely different things from the same first screen.

03 — Defining the Opportunity

Make the intelligence accountable

If the product's promise is intelligence, the design problem is not generating predictions — models do that. It is making a prediction something a driver can reasonably decide whether to believe.

01
Show the working
Every AI claim must carry the telemetry it was derived from, close enough to read without hunting.
02
Assume no network
Offline is the normal case, not the error case. The interface should never degrade into a spinner.
03
Collapse the app sprawl
One surface for car, charging, home energy and route, without becoming an unusable dashboard of everything.
04
Serve four segments at once
A single home screen that adapts rather than four separate modes.

04 — Design Strategy

Architecture first,
because the promise is structural.

A product whose value proposition is trustworthy intelligence cannot be retrofitted onto a network-first, monolithic client. The strategy was to establish the structural guarantees before building features on top of them — which is also why this project is honest about being unfinished.

01
Define the module boundaries
25 Gradle modules with convention plugins, so a feature module builds in four lines and cannot reach into another's internals.
02
Make offline the default
The data layer treats the network as an enhancement over a local source of truth, rather than a dependency with a cache bolted on.
03
Bind provenance to prediction
Every AI output carries its inputs by contract, so the UI cannot display a claim it cannot substantiate.
04
Build the design system early
Tokens, components and the 3D viewer completed before feature work, so the eleven modules render consistently by construction.
Back Return to the case study