BizPilot AI
A multi-agent business operations copilot that converts informal WhatsApp customer requests into verified GST quotations, invoices, and multilingual customer replies.
Project overview
BizPilot AI is a multi-agent business copilot designed for small service businesses that receive customer enquiries informally—through WhatsApp, phone notes or chat. The system converts those unstructured messages into professional quotations, invoices and multilingual replies through a five-agent transparent pipeline.
The problem
Small service businesses—caterers, tailors, repair shops—frequently receive requests like “mujhe 3 saree blouse chahiye, price kya hai?” and manually calculate pricing, GST, prepare a quotation PDF, and write a reply—every time. This creates delays, inconsistencies and errors under volume.
Intended users
- Small service business owners handling quote requests manually
- Solo operators who need to respond in the customer's language
- Businesses with returning customers who should receive loyalty discounts automatically
System pipeline
Intake Agent
Parses unstructured customer message, extracts intent and quantities.
Context Agent
Looks up customer history to determine if loyalty discount applies.
Generation Agent
Calculates pricing, GST and generates quotation and invoice content.
Approval Agent
Routes orders above ₹12,000 to owner for manual confirmation.
Review Agent
Cross-checks calculated prices against pricing rules before output.
✓ Verified output
PDF quotation or invoice + WhatsApp-ready prefilled message + multilingual reply
Note: WhatsApp sharing uses wa.me prefilled links. The product does not currently integrate directly with the WhatsApp Business API.
Key features
- Quote and invoice generation
- Pricing rule verification
- Approval routing for high-value orders
- Professional PDF export via jsPDF
- WhatsApp-ready sharing via wa.me prefilled links
- Multilingual replies (English, Kannada, Hindi)
- Returning-customer loyalty discount (5 % off)
- Rule-based fallback when Gemini API is unavailable
- Responsive mobile-first interface
Technical architecture
Frontend
- React 19 + Vite
- Tailwind CSS v4
- Lucide React icons
- jsPDF for document export
Backend
- Node.js + Express
- Google Generative AI (Gemini 2.5 Flash)
- Anthropic SDK (Claude) — secondary
- Rule-based fallback engine
Challenges & trade-offs
Keeping the pipeline visible. Hiding all AI reasoning behind a single button makes it harder for business owners to trust the output. The design exposes each agent's step so the owner can audit what happened.
Fallback reliability. If the Gemini API is down, the business should still be able to generate a quotation. A rule-based engine handles price calculation deterministically when the LLM is unavailable.
WhatsApp integration. Full WhatsApp Business API integration requires business verification. The current implementation uses wa.me prefilled links, which work immediately without registration.
Honest future improvements
- Integrate real WhatsApp Business API with proper business verification
- Add persistent customer database (currently session-scoped)
- Support voice-note transcription for businesses using audio messages
- Add more regional language support beyond Kannada, Hindi and English