Sixt Neo

✨ Inspiration

We wanted to rethink car rental from the ground up. Today’s experience still involves queues, uncertainty about the actual car, generic upselling, and a stressful pickup process. But everyone already has a messaging app in their pocket, so why not make car rental feel as simple as texting a friend? That idea became Sixt Neo: a conversational, AI-driven rental experience that feels human, fast, and personal.

🚗 What it does

Sixt Neo introduces a seamless end-to-end rental flow inside iMessage, powered by a multi-agent architecture. The Booking Agent helps customers choose and book the right car. The Upselling Agent analyzes persona, availability and discounts to suggest meaningful upgrades. The Pickup Agent shows the exact assigned vehicle, makes it blink, and unlocks it. The Broker Agent routes every message to the right agent based on context. The result is a single smooth conversation without apps, waiting or guesswork.

🛠️ How we built it

We implemented a multi-agent system with clear JSON schemas and strict role definitions. We used n8n as the orchestration layer for agent coordination and API execution. We integrated the hackaTUM Sixt API for booking creation, vehicle assignment, upgrade offers, protections, addons and remote car actions. We designed a natural iMessage chat UX with smooth language and contextual memory. We built a persona-sensitive upsell engine based on deal information. We created a pickup flow that blinks and unlocks the real Sixt demo car on campus.

🚧 Challenges we ran into

We initially wanted to use RCS to deliver Neo's messages, but there are unfortunately waiting times of multiple weeks in order to get registered as a RCS business. Also we needed prompts that were strict enough for reliable routing and flexible enough for natural conversation. The model sometimes hesitated to trigger real HTTP actions such as blink or unlock. Phase transitions were sensitive and required clear rules. Designing a shared JSON schema that all agents could rely on took several iterations.

🏆 Accomplishments that we’re proud of

We built a fully functioning end-to-end rental experience inside iMessage. The multi-agent architecture behaves predictably and handles messy real inputs. The real Sixt car blinks and unlocks based on natural language. Upselling feels personal and relevant instead of generic. We translated the playful and confident Sixt brand feeling into conversational interactions.

📚 What we learned

We learned how powerful LLM-driven agent systems can be when the structure is clear and how fragile they become without strict schemas. Prompt design matters more than code and clarity leads to reliability. Users respond very positively to conversational flows once friction is removed. Working with real APIs exposes real UX gaps and sparks creativity. Trust is essential for upselling to feel natural.

🚀 What’s next for Sixt Neo

We want to expand to RCS and Whatsapp. We plan to add payment flows, loyalty integration and ID verification. Persona modeling can be improved for even smarter upgrade suggestions. Richer generative UI elements can enhance booking.