Healthcare
Appointments, reports and preparation instructions answered in Bangla, with a hard boundary at clinical advice.
A hospital or diagnostic chain in Bangladesh runs a phone line that is almost permanently engaged, and the cost of that is not only service quality. It is appointments that never got booked, tests that were done wrong because preparation instructions were not understood, and patients who travel to collect a report that was not ready. These are logistics problems, and they are solvable in Bangla without going anywhere near clinical ground. The boundary matters more here than in any other sector we work in: the engine handles scheduling, readiness, instructions, location and billing questions, and it does not interpret symptoms, suggest a diagnosis, or advise on medication. Anything that edges toward clinical goes to a human, by design and not by prompt instruction.
The problem before the product.
An engaged line is a lost appointment
Patients who cannot get through do not queue. They call a different hospital. The loss is invisible in your systems because the booking never existed.
Preparation instructions do not survive the phone call
Fasting windows, timing and what to bring, delivered quickly in a second language at the end of a rushed call, produce tests that have to be repeated.
Report readiness generates endless repeat contacts
The same patient calls three times to ask whether a report is ready, which is a notification problem being absorbed as a call-centre problem.
Clinical safety is a genuine constraint
Any system taking health calls will be asked a symptom question. A system without a hard, structural boundary will eventually answer one.
Built against your
systems of record.
Booking, rescheduling and cancellation
Against live availability in your HIS, including doctor, department, branch and slot, with confirmation sent before the call ends and the slot released properly on cancellation.
Report readiness and delivery
Status read from your LIS or HIS, plus proactive outbound notification when a report becomes ready, which removes the repeat-call pattern instead of answering it faster.
Preparation instructions in Bangla, repeated on demand
Fasting, timing, documents and what to bring, from your approved instruction set, delivered in Bangla and repeated as many times as the patient asks without impatience.
A clinical boundary that is structural
Symptom, diagnosis, medication and dosage questions are routed to a human. This is enforced in the flow design rather than requested of the model, because a prompt instruction is not a safety control.
Appointment reminders and no-show reduction
Outbound reminders with the preparation instructions attached, and a simple confirm or reschedule path, which recovers slots that would otherwise be lost.
Your systems stay yours.
Nothing here is a replacement platform. The engine reads from and writes to what you already run.
- Hospital information systems and appointment modules
- Laboratory information systems
- Existing PBX and contact centre
- SMS and WhatsApp for confirmations and instructions
- Billing and package management
- bKash, Nagad and SSLCommerz for advance payment links
What it will not do.
The constraints are the product. A system that cannot be audited cannot be approved.
- Patient data is health data. Deployment runs on-premise or in your own tenancy, and PII is masked before any prompt leaves your boundary.
- No clinical advice, triage, diagnosis or medication guidance. The boundary is enforced in flow design, not in prompt text.
- Preparation and procedure instructions are served from your approved content set only.
- Append-only recording and transcripts, retention set by you.
- Consent captured and carried for outbound reminders.
Why this sector
works differently here.
The regulators, the systems and the operating reality that decide whether a deployment in this sector actually survives contact with your environment.
DGHS oversight and a mostly private delivery system
The Directorate General of Health Services regulates facilities, while a very large share of actual outpatient and diagnostic care is delivered privately and paid out of pocket. A patient who cannot get through to you is not waiting in a queue, they are calling the next chain on the list.
Out-of-pocket payment changes the questions
Because most care here is paid directly by the patient rather than through insurance, price and package questions arrive on the same call as clinical logistics. Those are answerable from your billing system, and they are among the most common reasons people call before booking.
Diagnostic chains run on preparation compliance
Fasting windows, timing and what to bring decide whether a test is usable. Delivered hurriedly in a second language at the end of a call, they produce repeat tests. In Bangla, repeatable on demand, they do not.
Patients travel for reports
People physically come to collect results, so a report that is not ready becomes a wasted journey and a complaint. Proactive outbound notification in Bangla removes both the journey and the three calls that preceded it.
Health data should not leave the country casually
Patient data is the most sensitive category we handle. On-premise or in-country deployment is the default assumption for this sector rather than an upgrade.
The first build, in about 30 days.
Appointment booking, rescheduling and cancellation against live HIS availability, plus report readiness lookup, for one branch or one department.
- Booking, rescheduling and cancellation against live availability
- Slot released properly on cancellation
- Report readiness read from your LIS or HIS
- Preparation instructions in Bangla from your approved set
- A structural clinical boundary with human routing
Scoping everything at once is how these projects fail. These come after the first build is answering real calls.
- Proactive report-ready outbound notification
- Appointment reminders and no-show recovery
- Billing, package and insurance queries
- Multi-branch rollout
If three of these are true, book the diagnostic.
Your appointment line is engaged for long stretches of the day.
Patients repeatedly call to ask whether a report is ready.
Tests get repeated because preparation instructions were misunderstood.
No-shows are a material share of booked slots.
You have multiple branches and patients cannot easily find the right one.
Will it give medical advice?
No, and not because we told it not to. Clinical intents are routed to a human in the flow itself. A model instructed to avoid a topic will still occasionally address it, so we do not rely on instruction for a safety boundary.
Can it book against real availability?
Yes, where your HIS exposes availability to integrate against. Booking into a system the engine cannot read is guesswork, and we would not ship it.
What about patient privacy?
On-premise or own-tenancy deployment, PII masked before any prompt leaves your boundary, and identity verification before any patient-specific detail is spoken.
A diagnostic, then a written scope.
Forty-five minutes on your call mix and your systems. You get the findings in writing inside 72 hours, and they are yours whether or not we work together.