She talks
She taps the mic and says what's wrong, the way she'd say it to her sister. If something important is missing, Aira asks one short follow-up — it doesn't reject her and make her start again.
Aira voice intake
That's the part everyone accepted as normal. She's in pain, possibly frightened, and the first thing the building asks her to do is find a pen and describe it in a second language. Aira just listens instead.
“सीने में तीन दिन से हल्का दर्द है, चलने पर बढ़ जाता है, रात को नींद नहीं आती…”
symptomchest discomfort
duration3 days
severitymoderate · worse on exertion
sleepdisturbed
route toGeneral Medicine
Why this matters more than it sounds
Ask a Tamil-speaking grandmother to write 'intermittent, radiating, worse on exertion' in English and you won't get a symptom — you'll get a shrug and a word she's sure enough of to spell. The chart records the shrug. The doctor works from the shrug.
She taps the mic and says what's wrong, the way she'd say it to her sister. If something important is missing, Aira asks one short follow-up — it doesn't reject her and make her start again.
Symptom, duration, severity, and context come out as clean clinical fields. Rambling is fine. People in pain ramble, and a form that punishes them for it is a badly designed form.
It routes to the right department with the structured intake attached, so the consultation starts at her chest pain rather than at 'so, what brings you in today?'.
What Aira is not
Intake AI is the easiest place in a hospital to quietly start diagnosing people. So we drew the line early and put it where you can see it.
Aira captures what she said — symptom, duration, severity, context — and routes it. It doesn't offer a diagnosis, and it doesn't rank how urgent she is instead of a clinician.
The structured summary lands in the visit labelled as what it is: the patient's own account, tidied up. It's never presented to the doctor as a finding, because it isn't one.
If Aira misheard, the doctor corrects the field in the consultation and moves on. The transcript stays attached, so a correction is never an argument about what she actually said.
Before · after
Capture
She writes three words on a clipboard, in a language she's guessing at
She says it out loud, in her own language, and it comes out structured
Entry
The desk retypes her form into the system, adding a second chance to get it wrong
Nobody retypes anything — the structured summary is already in the record
Context
The doctor opens the visit cold and asks her the same questions again
Her symptoms and routing are on screen before she sits down
The queue
Intake is a bottleneck at a desk, so it only happens once she's reached the desk
Intake happens on her phone, before or on arrival — the desk isn't in the path
Use the words a real patient would — messy, hedged, and out of order. We'll run it in a live walkthrough and show where the workflow is ready and where it still needs work.
Next step
An elderly patient, a thick accent, a language your last vendor didn't support, a story that starts in the middle. That's the demo we want to do.