AI Scribe listens to your consultation and generates a structured draft note mapped to your template — SOAP sections or any custom structure you use. It's designed to reduce documentation time, not replace clinical judgment.
Patient consent is required before using AI Scribe. This is both a legal and ethical requirement. Do not start the scribe without first confirming consent.
One scribe pipeline, everywhere
AI Scribe uses a single pipeline — live-streaming transcription via Deepgram — everywhere it appears in Qliva: the standalone note editor (Notes → new/existing note) and the split-screen telehealth consultation view both use the same Scribe bar / recording control and the same transcription pipeline.
Both entry points behave the same way: your microphone audio is streamed in real time to Deepgram's speech recognition service, a live transcript appears as you speak, and the transcript is mapped into a structured draft note in your template, which you review and edit before it's saved.
Recording consent
Consent is captured inside the scribe interface itself, at the point you start recording — not as a separate step anywhere on the appointment screen. Before recording begins, you'll see a consent checkbox ("Patient has verbally consented to AI transcription") inside the scribe bar / recording control. Confirming it is what starts the recording, and the confirmation is written to the scribe session, not to the appointment record.
- Obtain patient consent. Have the conversation with your patient before starting — explain that an AI tool will transcribe the session for note-taking purposes.
- Confirm consent in the scribe UI. Tick the in-scribe consent checkbox before recording starts. Recording won't begin until this is confirmed.
Live Deepgram streaming
Used from the note editor (app/notes/new, app/notes/[id]) and the telehealth Consultation view alike — every scribe entry point in Qliva.
- Open the consultation note for the appointment (or an ad hoc note), or the Consultation view during a telehealth call
- Locate the Scribe bar / recording control
- Confirm consent, then click Start Recording
Your microphone audio is streamed in real time to Deepgram's speech recognition service over a WebSocket, and a live transcript appears as you speak. The scribe distinguishes between practitioner and patient speech where possible.
Speak clearly and avoid crosstalk for best results. The scribe performs significantly better in quieter environments. Rooms with echo or background noise will produce less accurate transcripts.
When you click Stop, a review overlay appears with the AI-generated draft broken into your note sections. Read through each section, edit anything that's incorrect, missing, or unclear, then click Accept to insert the draft into your note. Accepting appends the AI content — it doesn't overwrite anything you've already typed. If you need to capture more of the consultation, you can start a fresh recording afterwards and repeat the review-and-accept step as many times as you need.
Always review the AI-generated draft before signing the note. AI transcription can misinterpret clinical terminology, medication names, and dosages. The draft is a starting point — clinical responsibility for the final note remains with you.
Privacy and data handling
Your microphone audio is streamed to Deepgram for real-time transcription; the resulting transcript is mapped into your note template. Only the resulting transcript and draft note content are retained and attached to the note for audit purposes — raw audio is not kept once transcription is complete.