BabelDesk

Inspiration 🌍

I’m from Trentino, a valley where tourism is one of the main sources of income and hotels welcome people from all over the world every year.

At this hackathon in Rome, the problem feels even more real: Rome — and Italy in general — lives through tourism, with guests arriving every day from many countries, cultures, and languages.

With BabelDesk, we want to help hotels communicate better, welcome guests more warmly, and make language barriers disappear from the front desk.

What it does 🏨

BabelDesk is a real-time voice translation assistant for hotel receptions.

It listens to both the guest and the receptionist, detects the language, translates the conversation live, and outputs both text and voice.

How we built it ⚙️

We used ElevenLabs for speech-to-text and text-to-speech, and a local vLLM translation pipeline with DiffusionGemma, chosen because it is extremely fast.

Challenges we ran into 🚧

The main challenges were audio chunking, streaming responses, latency, and supporting multiple languages smoothly.

Accomplishments that we’re proud of ✅

We built a working prototype that translates real conversations in real time and feels useful for a real hotel scenario.

What we learned 💡

Streaming audio is hard, but when it works, it makes interaction feel natural and powerful.

What’s next for BabelDesk 🚀

Next, we want to add hotel-specific RAG, automatic check-in support, and document detail extraction from passports or booking confirmations.

Built With

• elevenlabs
• fastapi
• vllm