- Languages answered on one phone line
- 3
- Test calls handled
- 68
- Automated tests passing
- 55
Who the client is
The client type is a city hotel in Prague, where guests and callers use Czech, English and German. The work was a pilot built and tested in September 2026. The aim was to see whether one phone receptionist could cover all three languages and work through the hotel's own booking tools, with staff able to see what it did. Many Lebanese hotels, agencies and clinics face the same pattern: a small front desk, a ringing phone and callers who switch languages.
What the work looked like before
Calls in three languages reached a front desk that was also checking guests in. At busy hours, calls went unanswered. Bookings, changes and messages depended on whoever happened to pick up the phone. A guest standing at the desk gets attention first, so the caller waits or gives up. Nobody had a single record of who called and what they asked for.
What I built
- Built a phone receptionist that answers in Czech, English or German.
- Added language detection, so the caller is never asked which language they want.
- Connected the receptionist to the hotel's own booking tools, so it books, changes and cancels rooms there instead of in a separate system.
- Set up message-taking for calls that need a person to follow up.
- Created a front-desk dashboard that shows every call and booking live.
- Wrote 55 automated tests, including a check where five callers try to book the last 2 rooms at once.
A caller dials the hotel's number and the receptionist picks up. It recognises the language from the caller's speech and continues in Czech, English or German. It then books, changes or cancels a room through the hotel's own booking tools, or takes a message. Each call and booking appears on the front-desk dashboard as it happens, so staff see what was done without listening to recordings. In testing, the system handled 68 test calls and all 55 automated tests passed. In the hardest test, five callers tried to book the last 2 rooms at once, and the result was exactly 2 bookings with no double booking. The stack was Vapi, Deepgram, Vercel and Twilio. Because this was a pilot, I make no claim about live call volumes or savings. The results show that the build works under test.
Tools it runs on
Vapi, Deepgram, Vercel, Twilio. Built in the client's own accounts and handed over running.
Measured outcome
| Metric | After |
|---|---|
| Languages answered on one phone line | 3 (Czech, English, German) |
| Test calls handled | 68 |
| Automated tests passing | 55 |
| Five callers booking the last 2 rooms at once | exactly 2 bookings, no double booking |
Timeline: Built and tested as a pilot in September 2026.
Where this fits
This is one of the systems described on AI solutions in Lebanon; the audit-first process is on the how an engagement works page.
Questions about this build
Can an AI receptionist answer hotel calls in more than one language?
Yes. In the Prague pilot, one phone line answered in Czech, English or German. It detected the caller's language without asking. It was tested with 68 test calls.
Can an AI receptionist double book a hotel room?
It was tested for exactly that. In the pilot, five callers tried to book the last 2 rooms at once, and the result was exactly 2 bookings with no double booking. The receptionist works through the hotel's own booking tools.
Is this system running in a real hotel today?
No. It was a pilot demo, built and tested in September 2026, not a live deployment. The hotel is kept anonymous, and I do not claim live results.
How much does an AI receptionist for a hotel cost?
Fixed-price projects start from $1,000. The final price is set after a short audit of your calls, languages and booking tools.