The product that exists today
You dictate the way you already speak. It writes the record.
Our first product is an AI medical scribe, built for the linguistic reality of an Egyptian clinic: Arabic, English, and both inside the same sentence, with medical terminology in each.
01 · In the room
It listens while the consultation happens.
You keep talking to the person in front of you, and the phone on the desk stays face down.
A consultation in Cairo is rarely conducted in one language. The history arrives in Egyptian Arabic, the findings in English, and the switch happens mid sentence without the doctor noticing. The scribe was built for that from the start rather than adapted to it afterwards.
What it does
- Dictate, type, or upload. Speak naturally, or upload audio, or photograph a report and let it read it.
- It asks about what is missing. Not a form to fill. A short set of follow up questions about the fields that were genuinely not stated.
- It never invents a field. If you did not say it, it does not appear. Missing is shown as missing.
- Documents, not just notes. Operative reports and discharge or visit summaries as finished PDFs, on the form the hospital actually asks for.
- Ask across your own cases. Which cases had this finding last quarter, and how many are due follow up.
- A clinician reviews and signs. Nothing is filed until you approve it.
- You need
- A device with a microphone and a working connection.
- You do not need
- A server or a GPU, if you take the hosted deployment. If you would rather run it on your own infrastructure, that option exists too.
- Languages
- Egyptian Arabic and English, mixed. Clinical output in English, with Arabic names transliterated.
Right hemicolectomy
- Case
- DEMO-0148 · fictional patient, 54, male
- Indication
- Caecal mass, biopsy confirmed adenocarcinoma.
- Findings
- Mobile caecal mass, no serosal involvement. Liver smooth on palpation. No peritoneal deposits.
- Procedure
- Midline laparotomy. Right hemicolectomy with side to side stapled ileocolic anastomosis. Haemostasis checked.
- Blood loss
- Not stated
- Closure
- Mass closure, skin clips. Drain to right paracolic gutter.
A field that was not dictated stays in the record, marked, and is never filled in from context.
02 · Accuracy
The number, and how it was measured.
On our internal benchmark the scribe hears about ninety five words in every hundred correctly, and fills about four fifths of the structured record straight from what was said. The benchmark runs on synthetic text to speech audio rather than clinical recordings, so we treat the figure as a floor measured in quiet conditions and expect real dictation to sit above it.
Real consented recordings collected during a pilot are what would upgrade that claim. When we have them we will publish the new number, better or worse.
The fields it does not hear, it leaves blank and points out, instead of filling them with something that reads well. A record you are about to sign has to be wrong in ways you can see.
03 · The limits
What the scribe does not do.
- It does not make clinical decisions. It drafts, asks and files. Clinical responsibility sits with the treating clinician, exactly as it did before any software was involved.
- It does not file anything on its own. Nothing becomes a record until a clinician has read it and signed it.
- It does not talk to your patients. Nothing patient facing is offered today, including booking and messaging.
- It does not replace your record system by default. In most practices it runs alongside what you already have.
Where it runs is a separate decision, and it is yours. Manara can be hosted by us, or deployed on servers your practice controls. The two deployments, side by side
Dictate a case and read what comes back.
There is a working demo on this site: speak a fictional case in Arabic, English or both. Or bring us a real one and we will run it with you, thirty minutes, with a doctor on our team. If the scribe is wrong for your specialty we will say so in the call rather than after the invoice.