art-sparrow

Tabibu is the Swahili word for doctor. It is a Flutter app, on Firebase, that carries a person from booking to result in Nairobi, in English and Kiswahili. A patient books a doctor; the doctor orders a test and sends it to a pathologist; the pathologist reads a slide and signs a report; the doctor discusses it and approves a plain-language explanation; the patient reads it. When a confirmed result is a notifiable disease, it becomes an anonymous count, and an authorised clinician approves a bilingual alert before any patient sees it. The problem: Kenya has about one pathologist for more than 725,000 people, and public hospitals that could do histology reported a median wait of about 14 days for a result (Brand et al., Journal of Global Oncology). Reading the slide is one stretch of that wait; the order, the hand-off, the report and its explanation are the rest, and those are what software can shorten. The rule that shapes every screen: AI prepares, a qualified person decides. The AI's first read of a slide is a labelled prompt for the pathologist, scored by them before they write a word, with its confidence capped at Low; a patient reads an explanation only after their doctor approves it; an alert is delivered only after an approver releases it. I measured it and publish the number: on 200 held-out tiles the pre-read was right 72% of the time at best and still missed about half of the tumour tiles, so it is never presented as a diagnosis.