Notes from two days in a mobile medical van
Gloves on, one charger, six hours without signal. Everything we assumed about the interface was wrong by the second hour.
We went expecting to validate a design. We came back with a list of things to delete.
The first correction was physical. The worker entering records wears gloves for most of the morning, which rules out small targets and any gesture requiring precision. The tablet lives on a shelf at chest height, in bright sun, and is shared between three people across a shift. Anything requiring a login per patient was never going to survive.
We went to validate a design. We came back with a list of things to delete.
The second was linguistic. Typing happens in Hindi, in a transliterated form nobody standardises. Two workers spell the same village three ways. An identity system that assumes clean text produces three patients where there is one. Voice entry solved more of this than any amount of validation logic.
The third was about time. There is no quiet moment in which to complete a record. Entry happens in fragments between patients, and it must survive being abandoned halfway through. Autosave is not a convenience here; it is the entire feature.
The last was connectivity, and it was the least interesting. Six hours offline is trivially handled once you accept it as the normal case rather than an error state. The interesting failures were all human, and none of them were visible from a meeting room in Delhi.
What we deleted after two days
Per-patient login. A twelve-field intake form, cut to five with the rest inferred or optional. A confirmation dialog on save. Free-text village entry, replaced by a picker built from the block list. An offline indicator, because offline was not an exception worth signalling.
Key points
- Gloves rule out small targets and precise gestures.
- Transliterated Hindi defeats identity matching; voice entry beats validation rules.
- Records are completed in fragments, so autosave is the core feature.
- Six hours offline is trivial once treated as the normal case, not an error.
What we added
Voice entry for names and symptoms. A resume-where-you-left-off queue visible on open. A duplicate-merge prompt that takes one tap. Larger touch targets throughout, sized for a gloved thumb rather than a design system. And a shift summary the team reads aloud at the end of the day, which turned out to be the feature they valued most.
Questions this raises
Why not just train the staff more?
Turnover makes training a recurring cost. Design that needs less training is cheaper than training that needs repeating.
Does voice entry work in a noisy camp?
Reasonably, with a close-talk microphone and on-device models. It still beats gloved typing by a wide margin.
How long was the field visit?
Two full shifts. We would now argue for three, including one at a site with worse conditions than average.