An AI assistant that turns Italy's most confusing procedures into a 15-minute task. I designed the product end-to-end — conversational logic, information architecture and UI — and prototyped the front-end in working code.
Visit the live product — praticheeasy.it ↗In Italy, doing anything official — cancelling the RAI licence fee, claiming a bonus, registering a pet, renewing a document — means digging through fragmented, jargon-heavy government sites that assume you already know the answer.
PraticheEasy replaces that maze with one product: search a procedure in plain language, get a step-by-step guide personalized to your region and situation, and ask the AI assistant anything you're still unsure about.
I took it from a blank page to a working product — positioning, IA, the full interface and design system, the assistant's conversational logic, and the front-end itself, prototyped in code rather than static mockups.
Bureaucracy is stressful because it's opaque, generic and high-stakes. Every design decision came back to removing one of those.
People don't know a procedure's official name. Search had to accept how they actually talk — "disdire Sky", "bollo auto" — and route them to the right guide.
Deadlines and benefits change by region and situation. A short onboarding tailors guides and alerts, so nobody wades through rules that don't apply.
A wrong deadline costs real money. The assistant had to feel accurate and honest about its limits — never a confident wrong answer.
The whole product had to feel warm and reassuring, not like another cold government portal.
Most-searched guides — scannable cards with difficulty, category and time-to-complete
An LLM that sounds fluent but gets a legal deadline wrong is worse than no assistant at all. The real design work wasn't the chat bubble — it was the behaviour behind it.
Framed the assistant around Italian bureaucracy only, so answers stay relevant and defensible instead of drifting into general chat.
Engineered the system prompt and grounding so responses lean on real procedures rather than the model's assumptions.
Explicit "I'm not sure" and failure states, with a clear route back to the verified step-by-step guide.
Warm copy that lowers anxiety around a stressful task — trust is a feeling as much as a fact.
The AI is only as good as the failure states you design around it.
Guided journeys & deadline alerts — the product remembers so you don't have to