R x

The AI assistant

You practise the medicine. It does the typing.

The assistant in Nirnoy Rx does not write prescriptions, does not diagnose, and does nothing at all until you press something. What it does is take away the writing — and it tells you what that will cost before it does it.

Your signature, your decision

Nothing the assistant offers writes itself onto the prescription. Unless you tap it in, it stays on the screen.

Switch it off and nothing breaks

Prescriptions, patient files, the drug register, the desk — all of it works with the assistant off. It is a convenience, not a condition.

The price comes first

Beside every button is what it is likely to cost. You never find out from a bill.

Six jobs

What it does — and where each one stops

Reads a lab report

Photograph it; the values land in the file.

Tidies what you typed

Sorts a hurried note into complaints, examination and diagnosis.

Reminds you what you might reach for

Generic names with a reason — a list, never a prescription.

Reads back what you wrote

A second pair of eyes for what is missing.

Writes the advice in Bangla

In words a patient can take home and follow.

Shortens a long history

Many visits on one page, each line citing the visit it came from.

Lab reports

Photograph the report; the values land in the file

A CBC, a creatinine, an HbA1c. Photograph it and what is printed goes into the file exactly as printed — the figure, the unit, and that laboratory’s own reference range. Next time you can set it beside the last one instead of hunting for the paper.

  • It does not interpret. Only the rows the report itself marked stay marked.
  • It does not convert units. Two laboratories’ figures merged into one scale is a comparison that was never true.
  • What it could not read, it names — rather than guessing it.
  • It sets last time’s figure beside this one, and marks anything outside the range the report itself printed — arithmetic, not a reading.
  • The picture is not kept, the values are. You already have the paper.

The patient’s history

Twelve visits, on one page

For a patient who has been many times, the assistant writes one page: what is ongoing, what settled, which medicines came back. Every line names the visits it came from, so you do not have to trust it — you can go and look.

  • Allergies and current medicines are never summarised. They are printed verbatim beside it, from the file.
  • What the file does not contain is said plainly, rather than passed over.
  • Under three visits it does not offer to summarise at all.
1 2 3

What it costs

Priced in taka, before you press

Two ways. Put in your own API key and you pay us nothing — you pay the provider directly. Or buy credits from us. Either way, what a press is likely to cost is written beside the button before you make it.

  • One to five taka for most jobs; about two to five to read a lab report.
  • What each press actually cost is recorded in the software, job by job.
  • If something will be expensive it says so first rather than quietly spending it.
৳1–3 ৳

The boundary

What was deliberately not given to a model

Where being wrong hurts a patient, it is ordinary code: the same answer every time, with no internet and no cost.

The allergy and interaction checks

Matched from the allergy in the patient’s own file against the WHO naming stem — “Penicillin” on file catches “Moxacil” on the prescription. The formulary’s 1,321 interactions come from the same place. No model involved.

Doses and daily maximums

Straight out of the Bangladesh National Formulary — 1,106 monographs. It says where each came from, and that it is a reference, not an instruction.

And the part that matters most: the check never says “no problems found”. Every time, it prints how many medicines it could look up and names the ones it could not — because a doctor who learns that silence means safe is worse off than one who knows nothing was looked at.

Questions

The ones doctors actually ask

Does the AI diagnose, or decide the medicine?
No. What goes on the prescription is your decision, your signature and your responsibility. What the assistant offers is a suggestion — shown separately on the screen, and nothing reaches the prescription until you tap it in.
Does it interpret the lab report it reads?
No, and that is the point. What is printed — the figure, the unit, and that laboratory’s own reference range — is copied exactly. It marks a row only where the report itself marked it H or L; it never decides for itself that a value is abnormal. It does not convert one laboratory’s units into another’s.
What if the photograph is blurred?
Whatever it could not read is named in its own panel and that row is left out rather than guessed. A tidy table with a row silently missing is worse than no table: there is no way to tell which figure was read and which was inferred.
Does patient information leave my computer?
Only when you press an AI button, and only as much as that job needs. Reading a lab report sends the picture but not the patient’s name or history. With the AI switched off nothing leaves at all, and the rest of the software works exactly as before.
How do I know what it costs?
Beside every button, before you press it, in taka — usually one to five. With your own API key you pay the provider directly and us nothing. Without one, you can buy credits from us.
Is the allergy warning also the AI?
No — deliberately. Allergies, the same molecule written twice, daily maximums and interactions are checked by ordinary code against what is in the file and the national formulary. It works with no internet, costs nothing, and gives the same answer every time. The thing that hurts somebody when it is wrong was not left to a model.

Try it for thirty days

Start with the assistant switched off if you like. Turn it on later with your own key, or with credits.