AI and OCR Disclosure
Transparency regarding machine learning in medicine intake
Core Safety Principle
Label scanning is assisted by AI and may contain errors. Always verify medicine details against the physical packaging before distribution. Automated models never make final dispensing or distribution decisions without human review.
1. Where AI is Used
When a donor or volunteer uploads a photograph of a medicine box, blister strip, or bottle, VitaMend uses an optical character recognition (OCR) pipeline assisted by Google Gemini multimodal models to read text directly from the image.
The model extracts the printed brand name, active generic salt, manufacturer name, batch or lot number, and expiration date.
2. Third-Party Model Provider
Packaging images submitted for scanning are securely transmitted to Google Gemini vision APIs. Only images submitted directly through the intake scan form are sent. No donor account passwords, government identity documents, or financial information are ever transmitted to the vision model.
3. Known Limitations of Machine Vision
Computer vision models can make mistakes. Factors that can cause OCR errors include:
- Low lighting, glares, or reflections on foil blister packaging.
- Faded, curved, or partially obscured printed batch stamps.
- Non-standard date formats (e.g. DD/MM/YYYY vs. MM/YY).
- Similar-sounding drug names or alternate manufacturer spellings.
4. Mandatory Human Verification
Because pharmaceutical errors carry serious healthcare risks, VitaMend strictly treats automated OCR as an intake accelerator, not an authority:
- Donors review and correct every field before submitting their donation.
- Pharmacists and trained verification staff inspect physical packaging prior to inventory admission.
- Every physical shipment is cross-verified against verified packaging details before delivery to recipient clinics.
5. Questions or Feedback
If you have questions about our technology or wish to report an OCR discrepancy:
Email: contact@vitamend.in