ramzyq/traderwise

TraderWise is a voice-first business intelligence tool built for Ghana's informal market traders. Send a WhatsApp voice note in Twi, get a contextual decision-support response in seconds — powered by Claude AI and OpenAI Whisper. # 🛒 TraderWise > **Voice-First Business Co-Pilot for Ghana's Informal Market Traders.** TraderWise is a WhatsApp AI business co-pilot for traders in Ghana's informal markets (Makola, Kejetia, Agbogbloshie). Traders send a voice note in Twi; the system transcribes, translates, reasons through Claude, and replies on WhatsApp in seconds — never by deciding for them, but by helping them decide better. **Built for Ghana. Powered by AI. Grounded in Makola.** CBC Hackathon • University of Ghana • Track 3: Economic Empowerment --- ## Table of Contents - Why TraderWise - How It Works - Architecture - Project Structure - Tech Stack - Getting Started (Local) - Environment Variables - Production Deployment - Testing - Design Principles - Ethics Architecture - Business Model - License --- ## Why TraderWise Every morning over 100,000 traders in Ghana's markets wake up to make decisions that decide whether their families eat that week: how much stock to buy, who to extend credit to, how to price, how to avoid spoilage. They make these decisions **alone**, with no tools and no thinking partner. **TraderWise is the business partner they never had access to — until now.** In under 8 seconds, a trader sends a WhatsApp voice note in Twi and receives a contextual, thoughtful response that helps her think through — not decides for her. The goal: **reduce weekly spoilage loss for informal traders by 5%.** At 500 traders, that is **GHS 75,000** returned to families that earned it. --- ## How It Works ### The Voice Pipeline ``` Trader sends WhatsApp voice note (Twi / Ga) → WhatsApp Cloud API → FastAPI webhook (signature-verified, BSUID-aware) → Celery task queue (Redis broker) → worker → Audio: media ID → /transcribe → Whisper/Groq ASR → text + language → Text: pass through → Distress classifier + fraud checker (run BEFORE Claude — no reasoning LLM) → Load trader profile + last interactions → build system prompt (Ama persona) → LLM reasoning (pluggable: Claude / Groq / OpenA …