Meeting the clinic that exists

Most clinical AI is built for a clinic that does not exist in Bangladesh. The brief assumes a fifteen-minute consultation, a doctor alone with a patient who speaks for herself, an electronic record to write into, and an insurer deciding what gets paid. Every one of those assumptions is false here, and a product built on them does not underperform in Bangladesh. It simply does not fit.

The clinic that exists is different in every particular. The consultation is forty-eight seconds. The patient is accompanied by a relative who answers for her. The doctor writes on a paper pad and has never used an EHR. The record is a plastic bag. The patient pays cash, out of her own pocket, for everything. These are not problems to be fixed before the software can work. They are the conditions the software has to work inside, exactly as they are.

Building for the real clinic is not a compromise. It is the whole design discipline. Every choice in Glyph traces back to a fact about the encounter as it actually happens, and the ones that matter most are the ones a Western brief would never think to ask about.

The number that defines the Bangladeshi consultation is forty-eight seconds. The largest international review of consultation length ever assembled, covering one hundred seventy-nine studies and twenty-eight and a half million consultations, put Bangladesh at the bottom of the table: an average primary-care visit of roughly forty-eight seconds, against twenty-two and a half minutes in Sweden. A single doctor may see more than ninety patients in a day, in a country with one doctor for every fifteen hundred people.

You cannot take a history in forty-eight seconds. You cannot read three years of paper in forty-eight seconds. The visit is not failing at these things; there is simply no room in it to attempt them. So the design does not ask the visit to do what it cannot. It moves the history out of the forty-eight seconds entirely, into the time the patient spends waiting, and hands the doctor the result.

This is also why Glyph demands nothing of the doctor. A doctor running ninety consultations a day has no slack to learn software, and any product that asks for some of that slack has already lost. His only new actions are reading a card and tapping approve. Everything else happens around him, before and after the forty-eight seconds, so the scarcest resource in the system is the one thing the product refuses to spend.

In a Western consultation the patient speaks for herself, and an attendant in the room is unusual enough to note. In Bangladesh the accompanied visit is the default. An elderly woman comes with the son who manages her care. A patient who speaks a dialect comes with a relative who translates. A great deal of what the doctor hears about the patient does not come from the patient.

This breaks the core assumption of clinical AI built elsewhere, that the voice in the room is the patient's. Glyph is built the other way, around the attendant as the normal case. The first thing it resolves is who is holding the device. From there, every clinical fact is tagged with its source: patient-reported, attendant-reported, attendant-translated, attendant-observed. When the patient's account and the attendant's diverge, the difference is surfaced for the doctor rather than flattened into one story.

This is not a feature bolted onto a solo-encounter design. It is a different starting assumption, and it is the kind of thing that only gets built by someone who has watched the actual encounter, where the son answers before his mother can, and the doctor has to know which words were hers.

Bangladeshi doctors write on paper. Not because they are behind, but because the paper pad is faster than any software for a doctor with forty-eight seconds, and because the whole system around them, the pharmacy, the patient, the next clinic, runs on paper too. A product that begins by asking the doctor to abandon the pad has misread the room.

So Glyph does not replace the prescription. It augments it. The note is drafted in the format Bangladeshi medicine actually uses, CC, O/E, Ix, Rx, Advice, never the SOAP format of Western training unless the doctor explicitly asks for it. The doctor reviews what was drafted, edits it, and approves. The paper prescription still exists, still goes in the patient's hand. What changes is that when the doctor approves, the prescription is also signed with his cryptographic identity, which is the thing that makes it verifiable later at a pharmacy or a hospital.

The pad is not the enemy of the record. It is the record's oldest form, and the design works with it rather than against it: keep the paper the patient trusts, and add the signature the system needs. Fighting the pad would have cost the doctor's goodwill on day one. Signing it instead is how the paper becomes something more, without the doctor changing a single habit.

In Bangladesh, seventy-three percent of health spending comes straight out of the patient's pocket, the highest share in South Asia, and health costs push four and a half percent of the population into poverty every year. For most patients, the price of a test or a drug is not a billing detail handled by someone else. It is a decision about whether the family eats as well this month.

This makes cost a clinical variable, not an administrative one, and a tool that ignores it is giving advice the patient cannot afford to take. Glyph treats cost as part of the medicine: prefer the cheapest investigation that actually answers the question, and reach for the Bangladeshi brand and generic names the patient can find and afford, Napa rather than Tylenol, the local generic rather than the imported brand, not the Western defaults a foreign-trained model reaches for first.

None of this is the model deciding to be frugal on the patient's behalf. It is the model surfacing the options and their costs so the doctor, who knows this patient and this family, can make the call. In a system that runs on out-of-pocket cash, leaving cost out of the clinical picture is not neutral. It is a way of being wrong that happens to look tidy.

Put it together and a pattern shows. Every real design choice in Glyph is the shape of a fact about the encounter that actually happens. The history moves out of the visit because the visit is forty-eight seconds. The source-tagging exists because the attendant speaks. The note is paper-format and signed because the doctor writes on a pad and the system needs proof. The drug names are local and the tests are cheap because the patient pays cash. None of these came from a brief. They came from the chamber.

This is the difference between software built for Bangladesh and software ported to it. Ported software treats the country's conditions as friction to be overcome on the way to the design it already had. Software built here treats those conditions as the design. The clinic that exists, with its volume and its attendants and its paper and its poverty, is not the obstacle the product has to survive. It is the thing the product is for. Meet it as it is, or do not bother coming.