AI Receptionist and Voice Agent Development
I'm Nazmul Khalid, and I build custom AI receptionists: AI phone agents that answer every call, speak your callers' language, answer only from your real business facts, take booking requests and hand the call to a person when they can't help. The one I built for WebCode Technology answered real calls on its first day, replies in 1.5 to 2 seconds, stops talking 138 ms after a caller interrupts, and runs on a local phone line for about $0.03 a minute of AI. I work remotely with businesses worldwide.
Problems I solve
- Calls come in while staff are busy, at lunch or gone for the day, and those callers ring someone else.
- Off-the-shelf AI receptionists don't speak your callers' language well, or don't follow your booking rules.
- A bot that guesses a price or invents a service is worse than no answer at all.
- You can't see what each call costs, so you can't price the service or check your margins.
What you get
- An AI phone agent on your number: Twilio, or a SIP trunk through Asterisk from a local carrier
- A knowledge base built from your real facts, with a rule that it may state only those
- Booking requests with a reference number and SMS confirmation, confirmed by your staff
- Transfer to a person, and a clean hang-up when the call is done
- A portal for each business: calls, recordings, costs and the questions it couldn't answer
- Per-call cost tracking, deployment, and testing on real calls in your callers' language
What does an AI phone agent do on a call?
It picks up, greets the caller, answers questions from your facts, and either takes a booking request, transfers the call to a person, or ends it politely. It hears the caller and speaks back directly, so replies come in about 1.5 to 2 seconds, without the 2 to 3 extra seconds that a speech-to-text and text-to-speech chain adds to every turn.
It also behaves like a person on the phone. When the caller starts talking it goes quiet, because it tracks what the caller has actually heard, not what the AI has generated. Line noise at the start of a call doesn't restart the greeting. And when it doesn't know something, it says so and offers a person instead of guessing.
AI receptionist for medical offices and clinics
A clinic's missed call is usually a patient who books somewhere else. The receptionist I built answers a dental chamber's line in Bangla: it gives the chamber's real hours and days, takes a booking request with the patient's name and a callback number, and sends an SMS with a reference. The chamber confirms the time itself, so the AI never promises a slot it can't keep.
Its knowledge came from the chamber's own recorded calls: the questions patients really ask and the answers staff really give. I've also led healthcare builds beyond phone calls, including DocBee, a telemedicine platform, and BD Doctors Directory.
White-label AI receptionists for agencies
If you run an agency or software company, you can offer AI receptionists to your own clients under your name. The system I built already works this way: WebCode Technology runs it for several businesses, each with its own phone number, its own knowledge, its own portal login and its own cost per call, so you can quote each client and see your margin.
What does an AI receptionist cost to run?
In early measured calls, the AI cost about $0.03 a minute on Google Gemini Live, against about $0.17 a minute on OpenAI Realtime for the same agent. In Bangladesh, a 20-channel phone line and a local server added about $28 a month, and incoming calls were free. Your carrier and call lengths will change the details, so I measure every call from day one.
Proof: work I've shipped
- AI Receptionist (Voice Agent)
A custom AI receptionist that answers business phone calls, knows only the company's real facts and hands off to staff.
- DocBee
A telemedicine marketplace for real-time video consultations between patients and doctors.
- BD Doctors Directory
A doctor discovery platform with a built-in engine for onboarding doctors at scale.
I also deploy and run what I build. DevOps for startups →
Stack
- Node.js
- TypeScript
- Google Gemini Live
- OpenAI Realtime
- Twilio
- Asterisk
- SIP
- PostgreSQL
- Next.js
- Docker
How does a project with me work?
It starts with a call and a small paid plan, then the build runs in fixed-price milestones you can check yourself, and I deploy and support what I ship.
- 1
Call. A 15- or 30-minute call in your time zone to talk through what you're building and whether I'm the right fit.
- 2
Plan. MVP scoping ($400, about five days) for a new product, or an architecture review (from $750) for an existing one. You get a written plan and a fixed quote, and the plan is yours to keep. If you go ahead with the build, the scoping fee is credited against the first milestone.
- 3
Build. Fixed-price milestones, each one something you can click through and check. 30–50% upfront, the rest billed as milestones are delivered.
- 4
Launch. Production deployment with CI/CD, monitoring and backups, plus app-store release support if there are mobile apps.
- 5
Support. Optional maintenance from $750 a month, or server care from $150 a month per server.
Further reading
Questions
- How long does it take to build an AI receptionist?
- A first version can answer real calls within days. Mine took calls on its first day, and moving it to a local SIP line took one more day. Booking requests, SMS, a staff portal and cost tracking followed over the next two weeks. Testing each flow on real calls in your callers' language takes time too.
- Can an AI receptionist use my existing phone number?
- Yes. You can forward your existing number to the AI's line, or connect the AI to a SIP trunk from your carrier. The phone side sits behind one small interface, so moving from Twilio to a local SIP line needed no changes to the AI code. Some countries require the server to be local, which I handle.
- Which languages can an AI receptionist speak?
- Many languages, but each one has to be tested on real phone calls by native speakers. Mine answers in Bangla. On one AI provider the Bangla sounded Indian rather than Bangladeshi, so we switched to another that supports Bangladeshi Bangla directly. A demo in a quiet room won't show you those problems.
- Will an AI receptionist make things up?
- Not if it's built to refuse. Mine may only state facts from the business's knowledge base. When a caller asks for something that isn't there, it says so and offers a person. It also logs every question it couldn't answer, so the business can add the real answer instead of letting it guess.
- Can an AI receptionist book appointments?
- It can take a booking request: the caller's name, the day they want, and a number for the SMS confirmation, with a reference number read back. On my build the clinic confirms the actual time, because it knows the doctor's real schedule. A direct calendar booking is possible when your system has an API.
- What does it cost to build an AI receptionist with you?
- Each build is quoted after a short scoping, because the cost depends on your phone setup, integrations and languages, not the AI. Scoping starts from $400 and ends with a fixed quote. Running costs are separate: on my build, about $0.03 a minute of AI plus a phone line and server.
Start with a conversation
Tell me what you're building, and I'll reply with questions or a next step. See how we can work together.
Or send the details →