AI medical scribe in telehealth: automatic consultation notes in 4 languages

A doctor spends an average of 15 minutes per consultation on clinical assessment and 10 minutes on documentation. In telehealth — where the consultation is already time-constrained — that documentation burden is a real problem. The AI medical scribe in MedConnect eliminates it: real-time transcription during the consultation, automatic SOAP note generation, and a structured consultation report ready for the doctor's one-click validation.

This is not a future feature. It is deployed and operational across MedConnect installations in France, New Caledonia, Ivory Coast, and Saudi Arabia.

The documentation burden in telehealth

Every teleconsultation generates documentation requirements: the clinical assessment, the examination findings, the diagnosis, the treatment plan, the prescription. In a traditional workflow, the doctor types these notes manually — during or after the consultation. This creates two problems:

• Time: documentation takes as long as the clinical assessment itself. In a busy clinic doing 20–30 consultations per day, that is 3–5 hours of typing.
• Quality: notes written under time pressure are often incomplete, lack structure, or miss important details. A scribbled "chest pain, ECG normal, monitor" is not a useful clinical record.

An AI medical scribe for telehealth solves both problems simultaneously. The consultation is transcribed in real time, clinical entities are extracted automatically, and a structured SOAP note is generated before the consultation ends.

How the AI medical scribe works

MedConnect's scribe uses a two-stage AI pipeline:

Stage 1: Real-time transcription and entity extraction

During the consultation, the audio stream is processed by Azure Whisper — a speech-to-text model that handles medical terminology, accents, and conversational speech. The transcription runs in real time, with the text appearing on the doctor's screen as the conversation progresses.

Simultaneously, an Information Extraction Module identifies and tags clinical entities: symptoms, diagnoses, medications, vital signs, examination findings, and patient history. Each extracted entity is linked to its source in the transcript — so the doctor can verify any claim against the original conversation.

Stage 2: SOAP note generation

After the consultation (or at any point during it), the doctor can generate a structured SOAP note. The system organizes the extracted entities into the standard SOAP format:

• S (Subjective): patient's reported symptoms, medical history, reason for visit
• O (Objective): examination findings, vital signs, device data (ECG results, auscultation findings)
• A (Assessment): clinical assessment and diagnosis
• P (Plan): treatment plan, prescriptions, follow-up instructions

The note is presented to the doctor for review and validation. One click confirms the note, which is then saved to the patient record. The doctor can edit any section before validation.

Multilingual: FR, EN, AR, IT

The scribe operates in four languages: French, English, Arabic, and Italian. This is not simply a translation of the interface — the automatic consultation notes are generated natively in the language of the consultation. The transcription model handles each language's medical vocabulary, and the SOAP note generator follows language-specific clinical conventions.

For international deployments, this multilingual capability is essential. The Saudi Arabia deployment uses Arabic. French deployments use French. And the scribe can handle consultations where the doctor and nurse speak different languages — transcribing each speaker's language separately.

Local data processing

A critical concern for any AI medical scribe is data privacy. Where does the audio go? Who processes it? Where is the transcription stored?

MedConnect's architecture addresses this directly:

• Transcription processing can be configured to run within the deployment country — no audio data leaves national borders
• For on-premise deployments, all AI processing runs on local infrastructure
• Transcription data is stored in the same encrypted, access-controlled database as other patient data
• The scribe is activated manually by the doctor or nurse — it does not record by default

This is particularly important for government deployments where data sovereignty is non-negotiable. The same data residency guarantees that apply to the MedConnect platform apply to the AI scribe.

Elara AI agent: beyond the scribe

The scribe is one component of MedConnect's AI capabilities. Elara is a broader AI agent built on LangGraph/LangChain architecture with seven specialist sub-agents:

• Patient Specialist: search, create patients, OCR from ID cards
• Consultation Specialist: create consultations, search history, summarize cases
• Document Specialist: OCR, attach documents, extract structured data
• Knowledge Specialist: RAG (retrieval-augmented generation) over medical knowledge base
• Form Specialist: auto-fill structured clinical forms
• Staff Specialist: search staff, manage scheduling
• Admin Specialist: operational and licensing management

Elara accepts voice and text commands and adapts to the user's role (nurse, doctor, administrator). It can create a patient from an ID card scan, summarize a complex case across multiple consultations, or analyze uploaded documents — all within the same interface.

Accuracy and auditability

The AI scribe is a documentation assistant, not a diagnostic tool. Every generated note is presented to the doctor for review before it becomes part of the patient record. The doctor can:

• Edit any section of the generated SOAP note
• Click through to the source transcript to verify any extracted entity
• Reject the generated note entirely and write manually
• Customize the note template to match their clinical documentation standards

Observability is built in via Langfuse tracing — every AI operation is logged, traceable, and auditable. This is not a black box: clinical staff and compliance teams can review exactly what the AI produced and what the doctor validated.

FAQ

Does the AI scribe replace the doctor's clinical judgment?

No. The scribe generates a draft note from the consultation transcript. The doctor reviews, edits if needed, and validates with one click. The doctor retains full control over the clinical documentation.

What languages does the scribe support?

French, English, Arabic, and Italian. The transcription and note generation are native in each language, not translated.

Is the consultation audio stored?

The transcription text is stored in the patient record. Audio storage is configurable — it can be retained or discarded after transcription, depending on the deployment's data retention policy.

Can the scribe work in offline/async mode?

In asynchronous mode, the consultation is recorded locally and transcribed when processing capacity is available (either locally or upon sync with a server). The SOAP note is generated from the transcription and presented for doctor review.