DEV Community

VoiceFleet
VoiceFleet

Posted on • Originally published at voicefleet.ai

Building an AI Phone Agent for Irish B&Bs: A Technical Walkthrough

#ai #automation #hospitality #api

I recently helped an Irish B&B owner set up an AI receptionist. The technical implementation was surprisingly clean, and the results were immediate. Here's how it works under the hood.

The Problem (Technically)

A B&B phone line is a single point of failure. One line, one owner, zero redundancy. When the owner is cooking breakfast or showing a room, calls go to voicemail. Most callers hang up.

The technical requirement:

  • Intercept inbound calls before voicemail kicks in
  • Handle natural language conversation in multiple languages
  • Query availability in real-time (calendar API)
  • Book appointments / reservations programmatically
  • Forward urgent calls to the owner's mobile

Architecture Overview 

Caller → PSTN/SIP → VoiceFleet AI Engine → STT → LLM → TTS → Caller
                          ↕
                   Google Calendar API
                   Knowledge Base (FAQ)
                   Webhook → Owner notification
Enter fullscreen mode Exit fullscreen mode

The call flow:

  1. Caller dials the B&B number
  2. Call forwards to VoiceFleet's SIP endpoint (configured via carrier portal or simple call forwarding)
  3. AI picks up in <1 second
  4. Speech-to-text processes caller's speech in real-time
  5. LLM generates contextual response using the B&B's knowledge base
  6. Text-to-speech delivers the response
  7. If booking requested → Calendar API checks availability → confirms slot → sends SMS confirmation

The Calendar Integration

This is where it gets interesting. VoiceFleet's calendar sync uses OAuth2 with Google Calendar. The AI agent has read/write access to specific calendars.

When a caller says "Do you have a room available next Friday?", the flow is:

  1. NLU extracts: intent=check_availability, date=next_friday
  2. Calendar API query: GET /calendars/{id}/events?timeMin=...&timeMax=...
  3. If no conflicting events → "Yes, we have availability. Shall I book that for you?"
  4. On confirmation → POST /calendars/{id}/events with guest details
  5. Webhook fires → owner gets push notification

The whole exchange takes ~15 seconds.

Multilingual Handling

Irish tourism means German, French, Italian, Spanish, and Chinese callers. VoiceFleet handles this with automatic language detection in the first 2–3 seconds of speech.

The LLM switches context language without any caller input. No "press 2 for French" — it just works. The knowledge base content is either pre-translated or translated on-the-fly (the B&B owner writes FAQs in English; the system handles the rest).

Setup for Non-Technical Users

The B&B owner I worked with is not technical at all. Setup took 15 minutes:

  1. Create account on VoiceFleet dashboard
  2. Enter business details, FAQs, policies
  3. Connect Google Calendar via OAuth
  4. Set up call forwarding on their phone provider (Eir, in this case)
  5. Test with a real call

No code. No server. No SIP configuration (VoiceFleet handles the telephony layer).

Results After 30 Days

  • 142 calls handled by AI
  • 38 bookings made directly through the phone
  • 0 missed calls (vs ~60/month previously)
  • €99/mo cost (vs losing ~€4,500/mo in missed bookings)

For Developers: Why This Matters

If you're building for the hospitality vertical, phone-based booking is still massive — especially outside the US where WhatsApp and phone calls dominate over web booking for small accommodations.

VoiceFleet's API lets you embed this as a feature in your hospitality SaaS. Each property gets its own AI agent with custom knowledge base and calendar. The flat pricing model (€99/property/mo) makes unit economics predictable.

Worth exploring if you're in the proptech or hospitality-tech space.

Top comments (0)