Stellantis · Citroën

CARA

The hardest decision was agreeing we didn't need more screens.

CARA started as a one-line brief: build a voice-assisted interface for car owners, where the interface itself is the assistant, nothing more. The instinct on a project like this is to design more. The actual decision was designing less, and convincing everyone that was the feature.

The problem

An almost-blank brief, and a pull toward the familiar.

The brief was one line: a voice-assisted interface for car owners, where the interface alone acts as the assistant. There was no existing pattern to copy from, and no shortage of instinct to fill that gap with a conventional app structure, tabs, settings, a menu.

That instinct was the actual risk. A voice-first assistant in a car has a different scarce resource than a typical app: the driver's attention, not screen space. The problem wasn't designing the assistant, it was resisting the default shape of a digital product long enough to ask what this one actually needed.

Approach

Three decisions, in a project that ran on almost no method at all.

Unlike the AI-first work elsewhere in this portfolio, CARA's core design thinking used no HMHT and no simulated users, built the traditional way, with AI tools entering only once the structure was settled.

Decision 01 · Load-bearing

Committed to a single screen, no drill-down, voice carries the load.

The default assumption was that a car companion app needs the usual structure, sections, settings, a way to go deeper. Getting agreement that voice would carry almost all the functional weight was the actual design fight, reframing the brief from "design more screens" to "design the constraint."

Decision 02

Ideated manually before introducing any AI generation tool.

Early product thinking was worked out the traditional way, before Google Stitch was introduced later to help iterate on visual direction. Bringing AI-generated visuals in earlier would have anchored the team on surface variations instead of the harder interaction question.

Decision 03

Carried one interaction model across two very different surfaces.

CARA needed to feel like the same assistant whether reached through the owner's phone or the dashboard tablet. Rather than design two products stitched together by branding, the single-screen model was carried across both, adapted only for context.

Outcome

Shipped, and unveiled publicly in September 2025.

Launched, Sept 2025

Unveiled publicly with Citroën's Basalt X range in India. Citroën's own launch materials called it "India's first and only in-car companion."

Built for reuse

Designed to be configurable across Stellantis's global brand portfolio, not a one-off feature.

Two surfaces, one model

Ships identically in the owner's mobile app and the car's in-car infotainment tablet.

Reflection

The trade-off is real: a single-screen, voice-first interface puts enormous weight on voice recognition doing its job well, there's no menu to fall back on. That's the bet the product makes, and it shipped anyway. This one stands apart methodologically too, no HMHT, no simulated users, proof that AI-first isn't the only way to get to a strong answer.