Optical & Evidence Pipeline

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.

01

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.

Key Technical Safeguards:
  • Client-side optical barcode & DataMatrix decoding
  • Extraction of GTIN, lot/batch, and expiry strings
  • Accessible browser camera capture or manual text input
02

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.

Key Technical Safeguards:
  • Domain safety classifier prevents unsupported requests
  • Identification of brand, generic, and strength entities
  • Controlled fallback for requests requiring clinical doctors
03

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.

Key Technical Safeguards:
  • Structured DGDA-aligned local catalogue lookup
  • Live openFDA Drug Label API connector
  • Candidate match status (no physical authenticity claims)
04

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.

Key Technical Safeguards:
  • 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.