Home Solutions DeepMed Dictate
DeepMed DictateFor iPhone and Android

Say it once. Approve a document you can trace.

Dictate, or record the consultation with consent. DeepMed Dictate drafts the summary, note or letter, links each drafted statement to its source, and nothing is final until you approve it.

Book a guided walkthrough
Coming soon to the App Store and Google Play.
Illustration with a fictional patient: a draft with its Why is this here? sheet open.Illustration with a fictional patient: the Recording screen, and a draft with its Why is this here? sheet open.
  • Nothing final until the doctor approves
  • Dictation or consented conversation
  • 6 document types
  • PDF, JSON and FHIR R4 export

From voice to approved document, in four steps

The phone records and reviews. The server transcribes, drafts and checks. The doctor decides.

Swipe for steps 2 to 4
  1. Illustration · fictional patient
    01

    Confirm, then record

    Find the patient and confirm the patient and encounter before a word is recorded. The server refuses a recording whose patient and encounter do not match.

  2. Illustration · fictional patients
    02

    It uploads and drafts

    While the app is open and online, each recording uploads in checked chunks and resumes where it stopped. The server transcribes it, extracts the facts with their sources, then writes the draft.

  3. Illustration · fictional patient
    03

    Review what needs you

    Critical issues, such as a contradictory allergy status, hold approval until a doctor resolves or dismisses them. Warnings are shown for review and do not block.

  4. Illustration · fictional patient
    04

    Approve, then export

    Approval needs your attestation and the MRN typed again. An approved version never changes; export it as PDF, JSON or a FHIR R4 document bundle.

Under the hood: what happens to a recording
  • Each take is written to the phone in encrypted one-second frames, so if the app is closed at most about a second is lost, and the take is offered back at the next launch.
  • Uploads go in 1 MiB chunks, each with its own SHA-256, plus a whole-file check. The phone deletes its copy only after the server confirms a matching file.
  • The transcript is timestamped, and the transcription service's original output is kept unchanged.
  • Two AI stages: facts first, each with its source and exact words; then the document, written from supported facts only.
  • Checks written in code, not the model, remove unsupported text before the doctor sees the draft.
  • Correcting the transcript or a speaker creates a new transcript version, and drafts built on the old one get an issue that holds approval.

Every statement has a source

A doctor should not have to trust a model. Tap a statement in the draft to see the fact behind it, the exact words and who said them.

  1. 1 In the draft
    Medications
    Apixaban 5 mg twice daily, commenced
    Metoprolol 25 mg daily, unchanged
    Furosemide, dose Not documented
    The draft is written only from facts that carry their source.
  2. 2 Why is this here?
    Fact
    Apixaban 5 mg twice daily, commenced
    Source
    Transcript · 02:14 · Doctor
    Quote
    “…commenced on apixaban 5 mg twice daily for the new atrial fibrillation…”
    Play
    Each fact keeps its exact words. If the quote is not in the source, the fact is dropped.
  3. 3 In the recording
    02:09…and the rate is now controlled.
    02:14She was commenced on apixaban 5 mg twice daily for the new atrial fibrillation.
    Playing from 02:14
    Play the recording from that moment, through a link that expires.
Removed before you see itChecked against the source, not trusted
Warfarin, startedQuote not found in the source · fact excluded
Counselled on bleeding risk.No supporting fact · sentence removed
Illustration with a fictional patient: the Document tab, the Why is this here? sheet and the Transcript tab. The Removed panel explains checks that run before the draft is shown; it is not an app screen.

Checks that do not rely on the model

  • Negation, uncertainty (“?PE”, “possible”) and who said it are kept.
  • Medication doses and units must appear in the source; otherwise the field reads “Not documented”.
  • “No known allergies” only from an explicit statement; contradictory allergies are flagged critical.
  • Numbers that are not in the cited facts are flagged for review.
  • Instructions hidden in a recording or a document are neutralised, and no check relies on the model obeying.

Dictate alone, or record the consultation

Both modes confirm the patient first, and both drafts go through the same checks and the same approval.

Dictation

Dictate after the round or between patients. Add typed notes as you go, or append another recording later; drafts from the earlier transcript are flagged.

Conversation, with consent

Consent comes first: who gave it, their relationship, how, any witness and the policy version. Withdrawing it stops processing. Speakers start as Unknown, and you assign each one before anything is drafted. Recording a conversation needs a connection, so consent can be captured.

Six documents doctors actually write

Each type has its own sections, set per organisation in versioned templates. A medication detail that was not said reads “Not documented” instead of being guessed.

  • Discharge summary

    Medications with status and changes, pending results with who is responsible.

  • Consultation note

    From a dictation, or the recorded conversation itself.

  • SOAP note

    Each section written only from what was said or typed.

  • Progress note

    A dated update on the patient's course.

  • Referral letter

    The reason for referral, informed by the supporting documents you select.

  • Patient instructions

    Plain language, written only from the approved document, with its own review.

Export as PDFJSONFHIR R4 bundle Drafts carry a DRAFT label on every page, and every export is logged.

Attach PDF letters and results that contain text: the text is extracted, uncertain passages are flagged, and only the ones you select inform the draft.

Built for the ward, run by your organisation

What the app does when the signal drops, who controls it, and how patient data is protected.

On the ward

Works offline
Dictate and type notes for recently opened encounters. The take uploads once you are online with the app open.
Recovered, not lost
If the phone closes the app, at most about a second is lost, and the take is offered back at the next launch.
Interruptions handled
If another app takes the audio, recording pauses and asks you to resume. On Android it carries on with the screen off.

Your organisation

Versioned templates
Set per organisation; changing a template never rewrites a document that used it.
AI switched per practice
A system administrator enables each AI service, and AI processing is turned on or off for each practice.
Standard exports
PDF, JSON or a FHIR R4 document bundle; sharing confirms the patient and the destination. FHIR output is not yet validated against profiles.

Patient data

Only your patients
Every clinical action is checked against the organisation and the doctor's access to that patient.
AI only when switched on
No patient data leaves the server unless the AI service is enabled and the organisation allows AI processing.
Encrypted on the phone
Recordings, drafts and offline copies, with a key held in the phone's secure storage.

Where DeepMed Dictate fits

Dictate covers the first step: capture, then a document the doctor approves, ready for the platform.

  1. DeepMed DictateThis page

    Dictate and approve 

    Dictate or record the consultation with consent; the doctor approves the draft.

    Exports PDF, JSON or FHIR R4
  2. DeepMed Platform

    Read and protect 

    Scans become text, and identifiers are masked before any AI sees them.

  3. DeepMed Platform AI-Assisted Coding

    Code 

    Suggested codes arrive with the passage behind them, for the coder to accept.

  4. DeepMed Platform Coding Audit

    Audit 

    A second AI pass scores the coding and proposes changes for the auditor.

  5. DeepMed Platform CDI

    Clarify 

    Documentation gaps become neutral, non-leading physician queries.

  6. DeepMed Platform DRG Grouping

    Group 

    A rules-based grouper assigns the DRG, final when the chart is locked.

Use one step or all of them. An approved Dictate document exports as PDF, JSON or FHIR R4, and the platform reads PDFs, scans, Word files and text.

Connect it to your systems

Get DeepMed Dictate

Coming soon to iPhone and Android. See it now in a guided walkthrough, and ask how we would set up your organisation.

Coming soon to the App Store and Google Play.
Book a guided walkthrough
Top