Claims about bilingual AI are easy to make. The proof is in the transcript. What follows is a realistic call reconstruction — the kind of conversation that happens dozens of times a day on Maraba-powered pharmacy and clinic lines across Lagos. The transcript is annotated to show what the system detects at each step.
The business in this scenario is a pharmacy on Bode Thomas Street in Surulere. The knowledge base contains stock information, pricing, opening hours, and a rule that prescription queries requiring pharmacist judgement are escalated to the pharmacist's direct line.
The call
"Ẹ káàárọ̀ — good morning o — please, I want to ask about something."
System detection: Opening segment identified as Yoruba — Ẹ káàárọ̀ (formal morning greeting). Caller then switches to Nigerian English. Language profile: Yoruba-English bilingual. No menu required, no prompt needed. Maraba processes the full sentence.
"Ẹ káàárọ̀. Good morning! You're through to Bode Thomas Pharmacy. I'm Maraba — how can I help you today?"
What Maraba did: Matched the caller's greeting register — opened with the Yoruba greeting to signal that the language was understood, then switched to English for the main response body, mirroring the caller's own language pattern. This is not a scripted decision. It is a language modelling output that reflects the caller's demonstrated preference.
"Yes, please. So — ṣé ẹ ní — do you have Augmentin 625mg? The one in the orange pack."
System detection: Mid-sentence code-switch. Ṣé ẹ ní is Yoruba for "do you have." The caller inserted it as a natural bridge, then continued in English with the specific drug name. STT processes both segments sequentially — Yoruba model for the first clause, Nigerian English model for the remainder. The transcription reads: "Ṣé ẹ ní — do you have Augmentin 625mg? The one in the orange pack." The ṣ diacritic is preserved.
Intent classification: Drug availability query. Medication: Augmentin 625mg. The LLM queries the pharmacy knowledge base.
"Yes, we have Augmentin 625mg in stock — the GSK orange pack. We have it in strips of 6 tablets and boxes of 21. Is there anything else you need?"
"Okay good. And the price? Because last time it was ₦4,500 but my neighbour said it went up — ó pọ̀ síi — is it more now?"
System detection: Predominantly English with a Yoruba phrase inserted for emotional emphasis. Ó pọ̀ síi means "it has increased / it is more." The caller used Yoruba to express a concern — this is a common pattern where the mother tongue carries the emotional weight of a statement. The STT captures both segments. The intent is now a pricing query, with the subtext of price concern (the caller is price-sensitive and hoping the answer is no).
"The current price for a strip of 6 is ₦4,800, and a full box of 21 is ₦15,500. Prices did go up slightly in March — but those are our current rates. Would you like me to note your name so we can set one aside for you?"
What Maraba did: Acknowledged the price increase honestly (it is in the knowledge base as a March price revision). Did not deflect. Offered a practical next step — reserving stock — which serves both the caller and the pharmacy.
"Ah okay. E jọ̀ọ́, set one aside for me. My name is Funmi Adebayo. I'll come after work — around 6pm."
System detection: E jọ̀ọ́ is Yoruba for "please." Used here as a polite preamble to a request — again, the Yoruba phrase carries the politeness register while the English carries the content. Name captured: Funmi Adebayo. Intended visit time: 6pm today.
"Done, Funmi. I've noted your name and will flag this for the pharmacist. We're open until 8pm, so you have plenty of time. See you this evening."
"E ṣéun. Thank you so much."
System detection: E ṣéun is Yoruba for "thank you." The caller closed in Yoruba — returning to the mother tongue for the parting word, as many Yoruba speakers naturally do.
"E ṣéun, Funmi. Take care."
What Maraba did: Closed in Yoruba, matching the caller's register. A brief, warm ending.
What happened in that 58-second call
The caller switched languages four times across seven utterances. Yoruba greeting, English request, Yoruba bridge phrase, English drug name, English main sentence with Yoruba emotional insert, English request with Yoruba politeness marker, Yoruba closing. Every switch was handled without pause, without the caller being asked to repeat herself, without a single language error in the transcript.
The post-call summary delivered to the pharmacy owner's WhatsApp within 60 seconds read:
Caller: Funmi Adebayo
Intent: Drug availability + pricing query
Outcome: Resolved — stock confirmed, name noted for 6pm collection
Sentiment: Positive (mild price concern, satisfied with response)
Priority: Normal
Action: Hold one unit Augmentin 625mg GSK for Funmi Adebayo, 6pm today
What this means for your business
The pharmacist saw this summary on WhatsApp at 10am. She noted the hold. When Funmi arrived at 6:10pm, her medication was ready. No call was missed. No instruction was lost. No one had to remember a verbal conversation from eight hours earlier.
This is what bilingual AI call answering looks like in practice — not a demo with perfect pronunciation and a cooperative caller, but a real conversation with natural switching, emotional cues, and a specific action item at the end.
Maraba's Starter plan, at ₦20,000 per month, handles 200 calls like this one every month, in Yoruba, Hausa, Igbo, and English, with a WhatsApp summary within 60 seconds of every call.
Free plan available — 50 calls in English, limited beta spots. Upgrade to Starter for full Yoruba, Hausa, and Igbo support with code-switching.
Request beta →