RAHA — WhatsApp Giveaway Bot
A production WhatsApp chatbot with real-time AI fraud detection on submitted photos.
Problem
A consumer giveaway campaign needed a fully WhatsApp-native entry flow — data collection, product-code validation, photo submission — with real-time fraud screening and no multi-second cold-start lag breaking the chat experience.
Approach
Production WhatsApp chatbot built on Flask, using Meta's WhatsApp Cloud API for messaging and Convex as a serverless backend. Groq's Llama 4 Scout vision model does real-time fraud detection on submitted photos, flagging AI-generated or invalid images via prompt engineering. A web admin dashboard handles entry monitoring, bi-weekly winner draws, CSV export, and product-code management.
The number
Cold start cut from ~43s to under 2s (95%+ improvement) via Convex serverless caching + pooled connections.