🧾 KHIPU

AI-Powered Invoice Validation with Local Multimodal AI + Web3 Payments

AI-powered invoice reconciliation for LATAM SMEs. Runs locally. Pays on-chain.

---

🎯 The Problem

Small logistics companies in Latin America process hundreds of invoices per month manually. The result:

• 💸 $5K–$50K USD/year lost to duplicate payments, fake suppliers, and quantity mismatches
• ⏱️ 7 minutes per invoice of repetitive human validation
• 🔓 23% of invoices contain discrepancies that go undetected until it's too late

💡 The Solution: Khipu

Khipu (Quechua for "knot", the Incan accounting system) is an end-to-end invoice validation pipeline that:

1. Extracts data from any invoice photo — even crumpled, low-quality ones
2. Validates using QVAC multimodal AI running locally on the user's laptop (no cloud, no data leak)
3. Reconciles automatically against Purchase Orders and Delivery Guides
4. Decides via traffic-light system (🟢 Approve / 🟡 Review / 🔴 Block)
5. Pays with one click through MetaMask (USDT on Ethereum/Sepolia)

🚀 Quick Start

# Clone
git clone https://github.com/gonnnzaDev/khipu.git
cd khipu

# Backend
python -m venv venv
venv\Scripts\activate    # Windows
# source venv/bin/activate  # Linux/Mac
pip install -r requirements.txt
python -m uvicorn main:app --reload --port 8000

# Frontend (separate terminal)
cd frontend/kiphu-frontend
npm install
npm run dev