Ambient AI Scribe

The doctor talks. The note writes itself. In Arabic and English.

Ambient consult recording becomes a signed bilingual SOAP note in under a minute. Khaleeji-tuned. PDPL-compliant. Locked the moment it’s signed.

PDPL 2021DHA alignedKhaleeji + EN7-day audio retention
Recording · 02:47
Capturing

Mariam Al Suwaidi · 42

EID 784-1985-3829471-1 · MRN-014902

Consent acknowledged in Arabic at 14:02

Draft · being written

Subjective

Patient reports 3 days of sore throat, low-grade fever, dry cough.

يشكو المريض من التهاب الحلق منذ 3 أيام مع حمى خفيفة وسعال جاف.

Objective

T 37.6°C, BP 120/78, HR 88. Pharynx mildly erythematous.

درجة الحرارة 37.6، الضغط 120/78، البلعوم محتقن قليلاً.

Hear a real consult and read the note it produced.

In-room recording — clean audio, no phone line.

Loading transcript…

Patient and agent voices generated from real scriptsTap a scenario to switch

From the first word in the room to a signed note.

The full flow takes under a minute of doctor time after the consult ends.

01

Consent + record

Patient hears the consent prompt in their preferred language — Arabic, English, Hindi, Urdu, Tagalog — and acknowledges. Recording starts on the iPad in the room.

02

Transcribe + structure

Deepgram nova-2 multilingual STT handles AR + EN code-switching. Sonnet drafts a bilingual SOAP plus vitals, prescriptions, and the follow-up date in ISO format.

03

Review + sign

Doctor edits inline, signs once. The signed note locks. An encounter auto-creates and flows downstream into coding and the claim XML.

Built for the way UAE doctors actually consult.

Khaleeji Arabic, not MSA

Code-switching and dialect handled. The scribe knows that 'البرد' and 'common cold' refer to the same thing in the same sentence.

EN

Patient has had a sore throat for 3 days.

AR

المريض يعاني من ألم في الحلق من 3 أيام.

Vitals + prescriptions extracted

Numeric vitals land in a structured table. Drugs land in a structured prescription list with dose, frequency, duration. Both flow into the encounter.

BP132/84
HR76
BMI28.4
HbA1c7.2%
Rx · Metformin 1g · BD · 90d

Follow-up dates resolved

The agent turns natural-language dates ('in 2 weeks', 'في خلال أسبوعين') into ISO-8601 timestamps and queues a follow-up booking task automatically.

“in 2 weeks”

«في خلال أسبوعين»

2026-05-19T10:00:00Z

Audio purges in 7 days

The transcript stays. The audio doesn't. PDPL-aligned default that protects patients without losing the clinical record. Per-org retention configurable.

Audio · purges day 7
Transcript + note · retained
How we keep it accurate

The note runs an audit before you sign it.

Other scribes claim “world’s most accurate” and ship boilerplate notes. We ship a structured fact sheet with every note: each vital, drug, allergy, and diagnosis cites the transcript snippet it came from. A pure-function audit then runs against a curated UAE drug formulary, ICD-10 hints, and the ISMP error-prone-abbreviations list. Critical findings block sign-off.

  • Hallucination guardrail. Every vital and finding cites a transcript quote. Empty fields are fine; invented ones get flagged.
  • Drug + dose validation. Levothyroxine in mg → flagged (1000× error). Pediatric vaccine code on adult → flagged. ISMP do-not-use abbreviations → blocked.
  • Family-vs-patient catch. "His father had MI" never silently lands in the patient's PMH.
  • Bilingual one source. Arabic and English render from the same structured object. They can't disagree on numbers.
Pre-sign-off audit
2 critical3 warning1 info
Sign-off blocked until 2 critical findings are acknowledged.
  • Drug "lay sicks" not in formulary. Suggest furosemide (Lasix) — confirm before signing.

    “...add lay sicks 40 mg twice daily...”

    drug · DRUG_UNKNOWN

    Acknowledge
  • Levothyroxine dosed in mg; canonical units are mcg. Could be a 1000× error.

    “...levothyroxine 50 mg daily...”

    drug · DRUG_DOSE_UNIT_MISMATCH

    Acknowledge
  • PMH "myocardial infarction" cites a family-relation token but is marked assertion="present". Should be "family"?

    “...his father had an MI at 58...”

    consistency · FAMILY_MISATTRIBUTED_TO_PATIENT

    Acknowledge
  • BP 140/90 mentioned in objective prose but not in structured vitals — possible hallucination or unit drift.

    consistency · PROSE_NUMBER_NOT_IN_VITALS

    Acknowledge

Real audit panel from /dashboard/scribe/notes/[id]. Findings, quotes, and rule IDs match the production code.

Compared to dictation.

Dragon-style dictation never matched what UAE doctors actually need: bilingual notes, dialect awareness, structured downstream handoff.

CapabilityVoice dictationGeneric AI scribeBookSlate Health
Bilingual SOAP (EN + AR)Sometimes
Khaleeji dialect handled
Vitals + prescriptions structured
Auto-encounter creation
ISO follow-up date resolved
PDPL consent capture
7-day audio purge by defaultSometimes

Listen to a real consult and read the note it produced — in 6 minutes.