The account and going native

In the personal account, the driver has everything that belongs to them: a garage of vehicles with detailed cards, diagnostic history, bookings, payments, notifications and support. Importantly, this is not a responsive website but a native mobile app, built on Capacitor, and that is why its use of the phone is genuine rather than simulated: geolocation to find the nearest garages, push notifications for status changes, external links opening in an in-app browser, and careful handling of photos, including converting the heavy HEIC format from iPhones into a web-friendly one. The interface feels alive thanks to micro-animations built on Motion and Lottie, without ever feeling overloaded.

The management panel: four roles

The second half of the product is the operational back office, and it is organised by role. Rather than a single administrative panel, it consists of four distinct workspaces: the platform administrator, the garage, the technician and the valet, each with its own dashboard and its own set of screens suited to its own work. The administrator has access to clients, garages, technicians, valets, diagnostics, requests, invoices and configuration; the garage to requests and quotes, bookings, invoices and finances; the technician to interventions and earnings; the valet to missions and earnings. Altogether that amounts to dozens of screens that together cover the whole operational side of the service. Large lists are built on virtualised, sortable and filterable TanStack Table components, so that even long registries stay fast, file uploads and data run through GraphQL, and the interface itself is multilingual. The panel is built on a ready-made template, but its working screens were written for the concrete needs of the service.

Engineering and quality

Underneath, both halves rest on shared principles. Data flows through GraphQL and Apollo, part of it in real time over web sockets and subscriptions, so that statuses and lists update themselves. Forms are built on React Hook Form with Zod schema validation, state lives in Zustand, and the client interface runs on its own design system built over Radix and Tailwind. Everything is written in TypeScript, and that is no mere formality: it is the types that keep a large project from coming apart. And for an app handling payments and bookings, tests matter especially, including end-to-end runs of entire user journeys - diagnosis, the guest path with a later transfer of history into an account, and adding a vehicle.

Built with Claude Opus

And honestly, about the main difference of this project: I did not write it by hand, I built it entirely together with Claude Opus, versions 4.6 through 4.8. The scale did not shrink because of that: two apps, payments, real time, four roles, dozens of screens - all brought to a working product in a short span of time. Here, artificial intelligence serves as a powerful tool rather than an autopilot: framing the tasks, the architecture, the boundaries between modules, the types, the tests and the single design system all remain my responsibility. In essence, this is my answer to what serious development with AI looks like when an engineer still sits at the wheel, and to how quickly large, real products can be built without losing control over quality.