eddyelly/ngaoai

AI powered Mobile Money Scam Detector for East Africa # 🛡️ NgaoAI - AI-Powered Mobile Money Scam Detector > *Ngao = Shield in Swahili* NgaoAI detects mobile money scams targeting M-Pesa, Airtel Money, and Tigo Pesa users in East Africa. Built specifically for Swahili - the first of its kind. ## How it works 1. User pastes or forwards a suspicious SMS/WhatsApp message 2. NgaoAI analyzes it using a fine-tuned Swahili NLP model 3. Returns: **SCAM 🚨** or **SAFE ✅** + explanation in Swahili ## Channels - 🌐 Web app (PWA) - 📱 WhatsApp Bot - 📞 USSD (works on 2G feature phones - no internet needed) ## Tech Stack - **Model:** Fine-tuned AfriBERT (Hugging Face) - **Backend:** Python + FastAPI - **Frontend:** React (Next.js) - **USSD:** Africa's Talking API - **WhatsApp:** Twilio / WhatsApp Business API ## Project Structure ``` ngaoai/ ├── backend/ # FastAPI server + model inference ├── frontend/ # Next.js web app ├── model/ # Training scripts + model files ├── data/ │ ├── raw/ # Raw collected scam messages │ └── processed/ # Cleaned, labeled dataset ├── scripts/ # Data collection + utility scripts └── README.md ``` ## Setup ```bash # Backend cd backend pip install -r requirements.txt uvicorn main:app --reload # Frontend cd frontend npm install npm run dev ``` ## Author **Edward Elisha Musabila** - GitHub: @eddyelly - Web: edwardelisha.com