Mythox  /  Work  /  Case Study AI·02

Blue Dive:
one agent, दो भाषाएँ.

The same AI receptionist offers Hindi mid-call, switches instantly and completes the entire booking in fluent Hindi. Write-up of one real, unedited bilingual call.

ClientBlue Dive Restaurant
DeliverableBilingual Voice Agent
Agent“Anika”
LanguagesHindi · English
Call length2:12, unedited
StatusLive on a real line
01The Brief

Your customers don't all call in English.

For a business serving Indian customers, an English-only phone agent turns away most callers at hello. Bilingual human staffing doubles the cost. IVR menus — “press 2 for Hindi” — kill the warmth before the conversation starts.

The brief: one agent, one number, every caller served in the language they think in — without pressing a single button.

02The System

Language is presentation. Logic is one codebase.

The booking pipeline is identical to the English agent — same slots, same guardrails, same reservation tools. Three components make the mid-call switch possible:

Multilingual STTHindi, English & Hinglish in one audio stream
Language-aware brainActive language tracked as conversation state
Native Hindi TTSNatural prosody, not transliterated English

One word from the caller flips every subsequent turn into Hindi — while the booking slots already captured (date, guests) persist across the switch. Phone numbers are read back digit-by-digit in the caller's language, which is exactly where most multilingual bots silently corrupt data.

03The Call

Recording & annotated transcript.

Real call, screen-recorded live. Same number as the English case — +1 938-222-3954. The switch to Hindi happens at 0:27.
Anika · AIThank you for calling Blue Dive, this is Anika. How may I help you today?Starts identically to every call — no separate Hindi hotline.
CallerHey Anika, I wanted to book a table at your restaurant.
Anika · AIThank you for choosing our restaurant. Would you like to continue in Hindi or English?Language offered conversationally — a question, not a keypad menu.
Callerहिन्दी में बात करते हैं।“Let's talk in Hindi.”
Agent switches to Hindi — rest of the call runs in Hindi
Anika · AIबहुत बढ़िया, अब हम हिन्दी में बात करेंगे। आप हमारे रेस्टोरेंट में टेबल बुक करना चाहते हैं — तारीख़ और समय बताइए।“Wonderful, we'll continue in Hindi. You'd like to book a table — please share the date and time.”The booking intent captured in English persists across the switch — the agent doesn't start over.
Callerकल शाम आठ बजे, चार लोग।“Tomorrow at 8, four people.”Date, time and party size extracted from Hindi speech — the slot-filling logic is language-agnostic.
Anika · AIअपना नाम और मोबाइल नंबर बताइए।“Please share your name and mobile number.”
CallerMy name is Ishan — मेरा मोबाइल नंबर 930441188 है।Mixed Hindi-English — handled without re-configuration.
Anika · AIधन्यवाद। आपकी डिटेल्स दोहराती हूँ — कल शाम आठ बजे, चार मेहमान, नाम इशान, नंबर 9-3-0-4-4-1-1-8-8। सही है?“Thank you. Repeating your details — tomorrow 8 PM, four guests, name Ishan, number 930441188. Correct?”Digit-by-digit read-back in Hindi — the guardrail travels with the language.
Anika · AIमैं उपलब्धता चेक कर रही हूँ… आपकी बुकिंग कन्फ़र्म हो गई है। धन्यवाद!“Checking availability… your booking is confirmed. Thank you!”Same live availability tool as the English flow — confirmed and logged before hang-up.
04The Outcome

Every caller served the way they think.

  • Full booking completed in Hindi — captured, read back and confirmed, no human involved.
  • ~6 seconds to switch languages — one sentence from the caller, zero configuration.
  • Context survives the switch — details given in English aren't asked again in Hindi.
  • One number, one codebase — adding a language is a presentation change, not a rebuild.

Next case study · AI·03

The same brain, on WhatsApp — qualifying leads by text.

Read WhatsApp case study