मुख्य कंटेंट तक स्किप करें

AI assist

AI suggestions are drafts — you must review and sign

Nothing the AI produces is saved or signed automatically. Every suggested value is highlighted until you confirm it, and signing is blocked until you do. You are responsible for what you sign.

What it does​

FeatureWhereNotes
Voice dictationMicrophone button under every text box in clinical documentsHold to talk, release to stop. English (India) or Hindi.
AI draft — preview (R2)"AI draft" panel at the top of the document editorPaste or dictate the consultation; the AI suggests values for the template's fields; you review field by field.
Patient-friendly summarySigned documentsEnglish or Hindi, generated only from the signed document; edit before sharing.
ICD-10 suggestions — preview (R2)Diagnosis entryOnly codes from the built-in ICD-10 list; you confirm each code.
Preview (R2)

AI drafts and ICD-10 suggestions are preview features planned for release R2. They stay unavailable until the platform operator switches previews on (CLINICAL_AI_ENABLED plus CLINICAL_AI_EARLY_FEATURES), even if a workspace ticks them in its settings. Voice dictation and summaries are part of R1.

Turning it on (workspace owner)​

AI assist is off by default. To enable it:

  1. Make sure the AI Assist module is enabled in the clinical configuration (depends on your plan).
  2. Open Settings → AI Assist.
  3. Read the AI data-processing addendum and tick the acknowledgement (required once per addendum version).
  4. Tick Enable AI assist for this workspace and choose the features:
    • Browser dictation (free) — uses the browser's own speech recognition. Chrome and Edge send the audio to the browser vendor's servers; enable it only if that is acceptable for your clinic.
    • Server speech-to-text (Indian languages) — audio goes through our server to the configured speech provider.
    • AI drafts, Patient-friendly summaries, ICD-10 suggestions.
  5. Choose the default dictation language and Save.

The page also shows this month's usage (AI drafts, summaries, transcriptions, audio minutes, tokens) and the monthly allowance. Only staff with the AI assist permission (doctors/practitioners by default) can use the features.

Dictating into a field​

  1. Click into a clinical document.
  2. Press and hold the microphone button under the text box (or focus it and hold Space).
  3. Speak; release to stop. The text is added at the end of the field as a normal edit (autosave applies).

Drafting a note with AI​

  1. Open a draft document and expand AI draft.
  2. Paste the consultation transcript or hold Dictate. Leave out names, phone numbers and addresses.
  3. Click Generate draft.
  4. Review the table: current value vs suggested value. Empty fields are pre-selected; tick or untick fields or whole sections.
  5. Click Apply selected fields. Applied fields are highlighted "AI suggestion — review and confirm". Check each one, edit if needed and click Confirm on that field ("Confirm all" does not confirm AI suggestions). The highlights are saved with the draft — they survive a reload — and the document cannot be signed until every suggestion is confirmed.

Suggestions that do not fit the template (wrong type, unknown field, value not in the allowed list) are discarded and counted in the review.

Privacy​

  • AI is never used for restricted records, restricted patients or — unless the owner explicitly allows it — psychiatry & psychology records.
  • The server sends only what is needed: the transcript and the template structure for drafts; the signed clinical facts without patient identifiers for summaries.
  • Transcripts, audio, prompts and AI answers are not stored by Achal Healthcare and are never written to logs. Providers are configured for zero data retention and no model training, with India-region processing where available (see the subprocessor list).
  • Usage is metered per workspace for billing and fair-use limits.

Limitations​

  • Dictation quality depends on the microphone and background noise; always proofread.
  • Browser dictation works in Chrome and Edge; other browsers use server speech-to-text when enabled.
  • Dictation is offered only inside a clinical document, and never for restricted or sensitive (e.g. psychiatry) records.
  • Recordings are limited to 60 seconds per press.
  • AI needs a connection (it does not work in offline mode).
  • If the AI service is busy you can always continue typing manually.