Inspiration

We wanted to replace the rigid, form-based Sixt rental flow with a natural conversation - like talking to a helpful counter agent - so users can state their trip needs and get an instant, tailored booking.

What it does

An AI checkout assistant for Sixt. Users describe their trip by voice or text; the system extracts booking details (trip type, passengers, duration, luggage, budget) and recommends suitable cars and protection packages, turning vague requests into a structured configuration.

How we built it

React + TypeScript frontend with TailwindCSS for a Sixt-branded UI. Google Gemini enables transcription and structured JSON extraction.

Challenges

Matching user requests to the features we actually support was a key challenge. We had to reliably translate messy natural language into valid options without over-interpreting or suggesting unavailable capabilities.

Accomplishments

A smooth Voice-to-Config flow where UI values snap into place after speaking. Built a resilient architecture that degrades gracefully on API hiccups.

What we learned

We learned that bridging the gap between natural language and strict application state is the hardest part of building AI interfaces. It's not just about getting text from speech; it's about reliably mapping a vague phrase like "I need a big car for my family" to specific JSON fields like { "passengers": 5, "luggage": "Heavy" }.