Concept buildA worked example of how we would build this for a typical business. Not client work: no client, no client data, no achieved results.
AI Voice Agents
Concept buildA concept build for a voice agent that answers a dental clinic’s phone after hours in Arabic and English and books routine visits into the clinic calendar. It sends an SMS confirmation, moves callers who describe urgent symptoms straight to the on-call dentist, and never gives clinical advice.
The brief
A typical dental clinic in the UAE closes its front desk in the evening, but patients keep calling. Some want to book a cleaning or move an appointment; a few are in pain and need to reach the dentist on call. After hours every one of those calls reaches voicemail, and the front desk spends the next morning returning messages, by which time many routine callers have booked somewhere else. Calls come in Arabic and English, sometimes both in one sentence, so a simple phone menu serves neither group well. The clinic needs the routine calls handled and the urgent ones escalated, without anyone, human or machine, pretending to be a clinician.
Explainer
Explainer · 7 slides · about 46 s
Concept buildConcept build: After-hours Arabic and English Call Line for a Dental Clinic
Concept build, explained in 7 slides: a concept build for a voice agent that answers a dental clinic’s phone after hours in Arabic and English and books routine visits into the clinic calendar.
Slide 1 of 7
01 · AI voice agents
After-hours Arabic and English Call Line for a Dental Clinic
Concept build: After-hours Arabic and English Call Line for a Dental Clinic.
02 · The problem
What a typical dental clinic is up against
A typical dental clinic in the UAE closes its front desk in the evening, but patients keep calling.
03 · The approach
How we would build it, in 5 steps
We would build it in 5 steps, from call mapping and the escalation script to privacy defaults and a supervised launch.
04 · Architecture
How the pieces connect
How the parts connect, from the first input to the person who reviews the result.
05 · Demo · sample run
A cleaning booked after hours, then an urgent caller
A sample run, step by step, including where a person steps in.
06 · Typical targets
What we would measure it against
Typical targets, each with the basis it would be measured against.
07 · Next step
Scope a build like this for your business
Book a strategy call to scope a build like this.
Architecture
Architecture of the After-hours Arabic and English Call Line for a Dental Clinic concept: Caller → Twilio phone line (call); Twilio phone line → Speech in and out (audio); Speech in and out → Red-flag rules (transcript); Red-flag rules → On-call dentist (urgent: transfer); Red-flag rules → Language model (routine); Language model → Clinic calendar (book slot); Language model → Clinic inbox (call summary).
Diagram with 8 parts: Caller (source), Twilio phone line (source), Speech in and out (processing), Red-flag rules (processing), Language model (model), On-call dentist (human), Clinic calendar (system of record), Clinic inbox (system of record).
Connections:
Sample run
Greeting in Arabic and EnglishRoutine
The caller answers in Arabic and asks for a cleaning next week.
Open slots offeredRoutine
Two cleaning slots read out from the calendar; the caller picks Tuesday 10:30.
Read-back and bookingRoutine
Name, mobile number and time read back, then saved through the calendar API.
SMS confirmation sentRoutine
Sent under the clinic’s sender name, with a reply option to reschedule.
Next caller: severe swellingNeeds care
A red-flag phrase matches before the model runs; the booking flow is skipped.
Transfer to the on-call dentistA person takes over
Connected with a spoken summary and logged for the morning team.
Build plan
We go through the clinic’s after-hours call log and the front desk’s call-back notes to list why people call: new bookings, reschedules, cancellations, directions and opening hours, and pain or injury. The clinical lead writes the escalation script with us: which phrases send the call to the on-call dentist, which mean the caller is told to contact emergency services, and what the agent must never say.
The phone line connects through Twilio to streaming speech-to-text and text-to-speech, with a language model in the middle that keeps replies short and plain. The agent greets in both languages and continues in whichever one the caller uses. Arabic and English are each tested with clinic staff playing callers, including Gulf Arabic phrasing and callers who switch language mid-call.
Every transcript passes a rules layer first. Phrases agreed with the clinical lead, such as severe swelling, heavy bleeding or a knocked-out tooth, skip the booking flow and transfer the call to the on-call dentist with a short spoken summary. The language model can raise a flag too, but it can never lower one, and it is never asked to judge symptoms.
The agent books only into slot types the clinic has opened for phone booking, through the calendar or practice system’s API, and reads the slot back before confirming. A confirmation SMS goes out under the clinic’s registered sender name, and the morning team finds a summary of every after-hours call in the clinic inbox.
Calls are not recorded by default; the clinic keeps only the call summary, in its own account, for a retention period it sets. Hosting sits in a region the clinic’s compliance lead approves, because UAE health-data rules can require patient data to stay in the country. Launch starts with one or two nights a week while staff read every summary.
Typical targets
Typical target
After-hours calls answered by the agent instead of going to voicemail
Basis: Capability: the agent picks up whenever after-hours routing is on; voicemail is only the outage fallback
Typical target
A routine booking completed in under 3 minutes of call time
Basis: The four-question booking script is timed on staff test calls in both languages during the pilot
Typical target
Every red-flag call put through to the on-call dentist’s line
Basis: Design property: red-flag rules run before the model, which can add a flag but never remove one
Typical target
A summary in the clinic inbox for every after-hours call
Basis: Design property: a summary is written at the end of each call, including transfers and hang-ups
Your project
The opening line, recording notice and opt-out wording checked for the places your callers are.
Numbers, call routing and the fallback to your staff set up under your accounts, with after-hours rules you choose.
Transcripts sampled and reviewed with your team, and every misheard phrase added to the test set.
Latency, accents and noisy-line behaviour checked on your numbers before callers reach the agent.
Where this fits
FAQ
No. The receptionist is a concept build, written to show how the call flow, red-flag rules and booking would fit together for a typical UAE clinic. The calls in the demo are role-play examples. Before any real launch, the clinic’s own clinical lead would write the escalation list and sign off the test calls in both languages.
It does not assess symptoms. If the caller mentions anything on the clinic’s red-flag list, the agent says it is connecting them to the dentist on call and transfers the call with a short summary. If the dentist does not pick up, it plays the clinic’s agreed fallback message, which tells callers with severe symptoms to contact emergency services.
That is common, so it is part of the acceptance tests rather than an edge case. Recognition quality for mixed speech varies between speech providers, so we choose one by testing it on staff role-play calls before the build is final. Names, dates and times are always read back before booking, which catches most recognition slips before they reach the calendar.
Yes. The clinic opens specific slot types for phone booking, such as check-ups, cleanings and follow-ups, and the agent can offer only those. New-patient consultations or treatments that need a clinician’s review can be set to request-only, so the agent takes the details and the front desk confirms the next morning.
The agent never promises a slot it could not save. If the calendar API does not confirm the booking, the caller hears that the clinic will call back to confirm, and the request appears in the morning summary marked as unconfirmed. If the agent itself is unavailable, the line falls back to the clinic’s existing voicemail.
Worked examples
Concept builds are worked examples, not client work.
Concept buildAI Voice Agents
A concept build for an outbound voice agent that makes routine ETA check-in calls to truck drivers and logs each answer in the fleet’s TMS. It calls only when the truck is parked or at a check-in window the driver chose, and it alerts a dispatcher straight away when a driver reports a breakdown or a delay.
Next step
Book a free 30-minute strategy call. Bring one process you want to automate and we will talk through scope, integrations and a sensible first step.
Free 30-minute call · no commitment
