AI WhatsApp Chatbot — Meera (Hey Concrete)
AI-powered sales assistant with self-learning capability and lead scoring system
Built a production-ready AI WhatsApp chatbot designed to act as a human sales consultant for Hey Concrete — a premium wall cladding brand.
The system goes beyond a basic chatbot by combining structured conversation design, lead qualification, and a real-time self-learning mechanism that improves responses without redeployment.
🚀 Core Features
1. Human-like Conversational Flow
- Designed natural WhatsApp-style interaction (short messages, emojis, guided flow)
- Collects user data step-by-step (name, location, budget, area, style, timeline)
- Avoids form-like experience — feels like chatting with a real sales agent
2. Lead Scoring Engine (100-point system)
- Budget alignment, area clarity, design intent, and engagement quality
- Automatically qualifies leads
- Triggers handoff to human sales (Kabir) when threshold is reached
3. Self-Learning System (Key Feature)
- Admin can correct chatbot responses in real-time
- Corrections are stored as “learning rules”
- Bot behavior updates instantly without redeployment
- Includes versioning + rollback capability
4. Knowledge Base Architecture
- Structured product data (wall panels, breeze blocks, murals)
- Includes pricing, dimensions, and product images
- Ensures zero hallucination — bot only responds from verified data
5. WhatsApp-Ready Architecture
- Designed for Gupshup API integration
- Supports quick replies, structured responses, and media (images)
- Modular backend allows easy deployment to messaging platforms
6. Follow-up Automation System
- Multi-layer follow-up strategy (20 mins → 4 hours → 24 hours → 3 days → close)
- Context-aware nudges (not generic “Hi” messages)
- Lead lifecycle tracking (active → dormant → cold)
🏗️ System Design
The chatbot is built as a modular system:
- Conversation Engine → handles flow + context
- Knowledge Base → product + FAQ retrieval
- Learning Layer → stores admin corrections
- Lead Engine → scoring + qualification
- Messaging Layer → WhatsApp API integration
⚙️ Tech Stack
- Node.js / Backend logic
- n8n (automation workflows & orchestration)
- LLM APIs (for natural conversation)
- REST APIs & Webhooks
- MongoDB (data storage)
- Cloudinary (image handling)
💡 Key Insight
Most chatbots fail because they are either:
- Too scripted (rigid), or
- Too AI-driven (hallucinate)
This system solves both by combining: 👉 structured flow + controlled AI + real-time learning
📈 Outcome
- Converts conversations into qualified leads
- Reduces manual sales effort
- Continuously improves without developer intervention
This project reflects my approach to building systems: 👉 practical, scalable, and aligned with real business use cases
