Login Page
Retailer orders
AI chatbot
Dashboard
Customer dashboard

Biznova - DevPost Submission (Short Version)

Inspiration

I watched my neighborhood kirana store owner struggle with a paper ledger while serving customers. When I suggested accounting software, he said, "Those are for big companies. I just need something simple—like talking to a friend who remembers everything."

That's when I realized: 63 million Indian MSMEs contribute 30% to GDP, yet 95% use manual bookkeeping. They can't afford Tally/Zoho (₹18,000-54,000/year), don't speak English, and lack digital literacy. Worse, they don't know which products are profitable, when to restock, or how to price competitively—costing them 20-30% revenue annually.

They needed an AI business advisor, not just a recording tool.

What it does

Biznova is a voice-first, multilingual AI business management system for Indian MSMEs.

Key Features:

Voice Commands: "5 Pepsi बेचे ₹150 में" → Auto-records sale, deducts inventory, calculates profit
AI Predictions: "Your Pepsi stock will run out in 3 days. Order 50 units now."
Conversational Q&A: Ask "What's my profit this month?" in Hindi/Telugu/English
Smart Insights: Profit optimization, seasonal trends, slow-moving stock alerts
Affordable: ₹299/month (94% cheaper than competitors)
Multilingual: Works in English, Hindi, Telugu with code-mixed support

How we built it

Tech Stack:

Frontend: React, TailwindCSS, i18next, Recharts
Backend: Node.js, Express, MongoDB
AI: Google Gemini Pro (intent classification), Google Cloud STT (voice)

Architecture:

React (Voice UI) → Express API → MongoDB
                      ↕
            Google Gemini Pro + Cloud STT

Key Algorithms:

Voice flow: Audio → STT → Gemini Pro → Confirmation → Execute
Auto-sync: Order completion → Create sale → Deduct inventory → Calculate COGS (atomic transaction)
Predictive inventory: daysRemaining = currentStock / (last30DaysSales / 30)
Profit analysis: Flag items with margin < 10%, suggest optimal pricing

Challenges we ran into

Noisy Environments: Kirana stores are loud → Added noise cancellation, push-to-talk, text fallback (92% accuracy)
Code-Mixed Language: "5 Pepsi बेचे ₹150 में" → Trained Gemini Pro with 10,000+ examples
Missing Cost Data: 60% didn't know product costs → Used 70% heuristic, gradual education
AI Hallucinations: Gemini misinterpreted commands → Always show confirmations, 80% confidence threshold
User Trust: Skeptical of AI → Started with read-only analytics, gradual introduction (85% adoption)

Accomplishments that we're proud of

92% voice accuracy in real-world noisy conditions
True multilingual AI (not just translation—understands code-mixed input)
85%+ demand forecasting with just 30 days of data
Atomic auto-sync (6 database operations in one transaction)
Pilot success: 20 users, 2.5 hrs/day saved, 48% error reduction, 85% willing to pay
100% discovered pricing errors they didn't know existed
67% followed AI recommendations (pricing, inventory)
Complete MVP with frontend, backend, AI, multilingual support, analytics