VaniKhata (वाणीखाता) is an AI-powered merchant dashboard designed for the next billion users. In a landscape where small Kirana store owners find digital bookkeeping tedious, VaniKhata introduces a zero-typing interface.
By leveraging Gemini 2.0 Flash, the app transcribes multi-dialect audio (Hinglish, Kannada-English, Hindi), extracts structured order data, and intelligently flags credit (Udhaar) transactions—all from a simple voice recording.
Converts messy, real-world audio into structured orders instantly. Supports mixed dialects like Hinglish and Kanglish.
Automatically flags credit requests based on linguistic intent (e.g., "Iska paisa baad mein dungi", "Khate mein likh lo").
A high-fidelity, glanceable UI for tracking daily sales, pending credits, and trending items.
AI-driven business insights that suggest cross-selling opportunities (e.g., "Customer bought milk, suggest bread").
Real-time data sync using Neon (PostgreSQL) and Prisma ORM.
- Frontend: Next.js 15, Tailwind CSS, Shadcn UI
- Backend: Node.js, Express, Multer
- AI Model: Gemini 2.0 Flash (Multimodal Audio API)
- Database: Neon (PostgreSQL)
- ORM: Prisma v7.4.0
- Deployment: Vercel (Frontend) + Koyeb (Backend)
cd server
npm installEnvironment Variables (server/.env):
DATABASE_URL="your_neon_connection_string"
GEMINI_API_KEY="your_google_ai_studio_key"
PORT=8000Initialize Database:
npx prisma generate
npx prisma db push
npm startcd client
npm installEnvironment Variables (client/.env.local):
# Local Development
NEXT_PUBLIC_BACKEND_URL=http://localhost:8000
# Production (Koyeb)
# NEXT_PUBLIC_BACKEND_URL=https://your-app.koyeb.appRun App:
npm run devVaniKhata removes the "typing barrier." It allows a merchant to maintain a digital ledger at the speed of speech, ensuring they never miss a credit entry or a sales trend. It is built for the reality of the Indian marketplace—where business happens in conversation, not keystrokes.
Team: VaniKhata Focus: Multimodal Audio Understanding & Utility for Bharat.