Prescription Label OCR + Human-Verification Flow for Medication Autofill

[hoangsonww]

Summary

Add a medication-label OCR flow so users can snap or upload a prescription label image and prefill medication reminder fields when barcode/QR data is unavailable.

Motivation

SymptomSync already supports barcode scanning, but many labels are damaged, missing, or not machine-readable. OCR-based autofill would reduce manual typing and lower data-entry friction while still keeping users in control.

Proposed scope (MVP)

1. Label capture/input

• Allow image upload and camera capture from the medication creation/edit flow.

2. OCR extraction

• Extract structured candidates for:
  • medication name
  • strength
  • dosage instructions
  • frequency text
  • optional NDC / Rx number if present

3. Parsing + confidence scoring

• Normalize OCR text into medication form fields.
• Return per-field confidence (high, medium, low) to drive UX.

4. Human verification step (required)

• Show extracted values in editable form fields before save.
• Require explicit user confirmation; never auto-save to database directly from OCR output.

5. Provenance + observability

• Store extraction metadata (source=ocr_label, timestamp, confidence summary) for debugging and quality tracking.

Suggested technical approach

• Frontend:
  • Add "Scan Label" CTA in medication form.
  • Reuse existing medication form validation, but prefill values from OCR response.
• Backend:
  • Add /api/med/ocr endpoint that accepts image input and returns normalized fields + confidence.
  • Use a pluggable OCR service layer so provider changes do not affect UI contracts.
• Data:
  • Add optional metadata columns (or side table) for OCR provenance and confidence snapshots.

Acceptance criteria

• User can upload/capture a medication label image from medication form.
• OCR response prefills medication fields in editable state.
• User must review and confirm before medication is created/updated.
• Low-confidence fields are visibly flagged for manual review.
• OCR failures return actionable messages without blocking manual entry.
• Unit/integration tests cover OCR parsing, confidence mapping, and confirmation gating.

Out of scope (later)

• Full handwriting recognition for cursive notes.
• Direct pharmacy integrations/e-prescription import.

Suggested labels

enhancement, help wanted