Table of Contents
💊 Introduction: The Silent Epidemic Killing Millions
Counterfeit pharmaceuticals are not just a market problem—they are a public health emergency. Across sub-Saharan Africa, over 500,000 deaths annually are linked to fake drugs. In Nigeria alone:
- 267,000 deaths yearly from substandard malaria medication
- 169,000 child deaths from fake antibiotics used for pneumonia
- A crisis so severe that regulators report it rivals casualties from war
This is not just data—it’s real lives lost.
That’s why I built the Fake Detector App—an AI-powered mobile solution designed to empower everyday Nigerians to verify medications instantly.
🧠 The Vision: AI as a Public Health Guardian
The goal is simple but powerful:
Put a drug verification system in every citizen’s pocket
By combining cutting-edge AI with real-time regulatory data, this project aims to:
- Detect counterfeit drugs instantly
- Reduce preventable deaths
- Build trust in healthcare systems
- Enable informed decisions at the point of purchase
⚙️ How I Built It: A Multi-AI Hybrid System Powered by Amazon Nova Ai
This project leverages a resilient, multi-provider AI architecture, with Amazon Bedrock and Amazon Nova playing a critical role.
🔹 Core Architecture
- Frontend & Mobile: Next.js 14 + Capacitor (Android-ready)
- Backend: Node.js + Prisma + PostgreSQL
- Deployment: Vercel + Firebase integrations
🔹 The AI Brain: Multi-Provider Fallback System
To ensure reliability, I designed a custom AI Fallback Manager that orchestrates:
- Google Gemini 2.5 Flash
- Anthropic Claude 3.5
- Amazon Nova (via Bedrock)
👉 If one AI fails (quota, latency, OCR issue), another automatically takes over.
This ensures:
- ⚡ Faster response times
- 🛡️ Zero single-point failure
- 🔁 Continuous availability in critical scenarios
🔍 Intelligent Drug Verification System
1. 📷 Smart Scanning with OCR + Vision AI (Amazon Nova AI)
Users scan drug packaging using their phone camera. The system:
- Enhances image quality (lighting, sharpness)
- Extracts batch numbers, manufacturer info
- Uses AI vision models for better accuracy on curved/metallic surfaces
2. 🧬 Semantic Matching with Transformers
Using Transformers.js (all-MiniLM-L6-v2):
- Extracted text is compared against thousands of official drug alerts
- Detects matches even with:
- Misspellings
- Formatting inconsistencies
- Partial data
3. 🔄 Real-Time Regulatory Sync
A custom engine continuously monitors:
- NAFDAC recall lists
- Public safety alerts
This ensures the app always has an up-to-date “Watchlist” of dangerous drugs.
⚡ Why Amazon Nova Matters in This Project
The Amazon Nova models (via Bedrock) provide:
- 🧠 Advanced reasoning for ambiguous drug labels
- 🖼️ Vision capabilities for complex packaging
- ⚡ Low-latency inference for real-time scanning
- 🔒 Enterprise-grade reliability
Impact:
Amazon Nova acts as a critical fallback intelligence layer, ensuring that even in poor network or high-load conditions, users still receive accurate results.
🚧 Challenges & Solutions
❗ OCR Limitations on Medicine Packaging
- Problem: Curved bottles + reflective surfaces
- Solution: Built a custom image preprocessing pipeline
❗ Serverless AI Scaling
- Problem: Heavy ML models in serverless environments
- Solution: Optimized ONNX runtime + Webpack configs
❗ Critical System Failures (JSON.parse Error)
- Problem: Tesseract crashes breaking user flow
- Solution:
- Bypassed local OCR
- Prioritized cloud-based AI (Amazon Nova, Gemini, Claude)
📚 What I Learned
1. 🛡️ Resilience is Everything
A health-tech system must never fail silently. Always have:
- Backup AI
- Fallback logic
- Redundant pipelines
2. ⚖️ Speed vs Accuracy Tradeoff
In life-critical apps:
Accuracy must NEVER be sacrificed for speed.
3. 🤝 Trust is a Feature
Users must feel:
- Safe
- Confident
- Informed
Clear UI/UX and transparent results are essential.
🌍 Real-World Impact: Saving Lives at Scale
This project has the potential to:
- Reduce counterfeit drug circulation
- Save hundreds of thousands of lives annually
- Support healthcare workers and pharmacists
- Strengthen regulatory enforcement
🚀 Driving Adoption in Nigeria & Beyond
To ensure real-world usage:
- 📱 Mobile-first design (Android focus)
- 🌐 Lightweight + offline-tolerant architecture
- 🤝 Partnerships with pharmacies & NGOs
- 📢 Awareness campaigns on drug safety
Future plans include:
- SMS-based verification for low-end devices
- Integration with national health systems
- Expansion across Africa
🛠️ Tech Stack
- Next.js, React, TypeScript
- Node.js, Prisma, PostgreSQL
- Transformers.js
- Firebase, Vercel
- Amazon Bedrock (Nova AI)
- Google Gemini, Anthropic Claude
🔗 Try the App
💡 Final Thoughts
The Fake Detector App proves that AI is not just about innovation—it’s about impact.
By leveraging Amazon Nova and a resilient multi-AI architecture, this project transforms smartphones into life-saving tools.
In a world where fake drugs can kill, verification should be instant, accessible, and reliable.