How MediGuard AI Works
Explore the 4-stage evidence-grounded architecture built to evaluate queries, cross-reference available catalogue records, and provide structured medicine explanations.
Input & Barcode Decoding
Manual packaging input & optical barcode capture
Users submit medicine queries via conversation or scan packaging labels using their camera. The browser-based scanner decodes standard GS1 DataMatrix codes, QR codes, and linear barcodes to extract product identifiers, lot numbers, and expiry strings without sending video streams to external servers.
- Client-side optical barcode & DataMatrix decoding
- Extraction of GTIN, lot/batch, and expiry strings
- Accessible browser camera capture or manual text input
Domain Guardrails & Entity Classification
Request validation & medicine entity extraction
The backend evaluates incoming requests to verify they fall within supported pharmaceutical information boundaries. The system identifies brand names, active generic compounds, and dosage forms while refusing out-of-scope non-medical queries or clinical diagnosis requests.
- Domain safety classifier prevents unsupported requests
- Identification of brand, generic, and strength entities
- Controlled fallback for requests requiring clinical doctors
Multi-Source Evidence Coordination
DGDA-aligned catalogue & live openFDA queries
The Evidence Coordinator queries two complementary pharmaceutical repositories: the DGDA-aligned local catalogue for domestic market records and the live openFDA API for official drug labels. All retrieved records are treated as candidate references, never as definitive proof of physical medicine authenticity.
- Structured DGDA-aligned local catalogue lookup
- Live openFDA Drug Label API connector
- Candidate match status (no physical authenticity claims)
Evidence-Grounded Interpretation
Strict prompt contracts & verified citation references
Retrieved records are assembled into a structured prompt contract. Google Gemini summarizes active ingredients, indications, and general precautions based strictly on the retrieved evidence. Every citation is validated by the backend before the response is delivered.
- Evidence-first assembly reduces unsupported AI claims
- Server-minted citation validation & arbitrary link filtering
- Mandatory clinical consultation disclaimers on every output
Try the AI Assistant on your medicines
Identify packaging labels, search active ingredients, and get clear plain-language guidance.