DEV Community

VoiceFleet
VoiceFleet

Posted on • Originally published at voicefleet.ai

Building a Multilingual AI Phone System for Healthcare in Ireland

Ireland's 2022 Census revealed something striking: 612,000+ residents speak a language other than English or Irish at home. That's 12.2% of the population across 100+ languages.

For anyone building voice AI or telephony systems, this creates a fascinating technical challenge — and a real-world use case where multilingual NLP has immediate, measurable impact.

The Problem Space

Healthcare phone systems in Ireland are almost exclusively English-only. This means:

  • 122,000 Polish speakers navigate appointment booking in their second (or third) language
  • 79,000 Romanian speakers may avoid calling entirely
  • Emergency dental situations go unreported when patients can't communicate symptoms

Technical Approach

We built VoiceFleet to handle this with a pipeline that includes:

  1. Language detection in the first 2-3 seconds of a call
  2. Real-time switching to the appropriate language model
  3. Domain-specific vocabulary — dental/medical terminology varies wildly across languages
  4. Cultural context — appointment scheduling norms differ (e.g., Brazilian patients expect WhatsApp confirmation, Polish patients prefer SMS)

The hardest part wasn't the NLP — it was handling code-switching. Many bilingual speakers in Ireland mix English and their home language mid-sentence.

Results

Practices using multilingual AI reception see:

  • 15-20% increase in new patient registrations from non-English-speaking communities
  • 30% reduction in appointment no-shows (native-language reminders are far more effective)
  • Near-zero language-related complaints

What We Learned

  1. Don't just translate — localise. Irish-Portuguese speakers use Brazilian Portuguese, not European Portuguese.
  2. Accent handling matters more than vocabulary for Irish English speakers.
  3. Fallback to human should be seamless, not a "press 0 for operator" experience.

Curious about the technical architecture? Happy to dive deeper in the comments.


Full deep-dive on our blog: voicefleet.ai

Top comments (0)