WhatsApp AI Chatbot with Product Compatibility (Automotive)¶
Status: Live demo
Demo: autobot.robiriu-dev.my.id (dashboard at /admin)

Executive Summary¶
A WhatsApp AI chatbot for an automotive electronics business that answers product, warranty, replacement, shipping, COD and FAQ questions from a knowledge base, checks product compatibility by car model and recommends the right product with an upgrade suggestion, transfers premium brands to a human agent, and captures every lead into a dashboard. The knowledge base and compatibility rules are editable without code.
How It Works¶
- Knowledge-base answers — Gemini answers warranty, replacement, shipping, COD and FAQ questions strictly from the business knowledge base.
- Compatibility logic — when a customer names a car (brand and/or model), the assistant checks the catalogue's fitment rules, recommends the products that fit, quotes the price, and suggests an upgrade where one exists.
- Human handoff — premium brands (BMW, Mercedes, Audi, and similar) and explicit "talk to a human" requests are not answered with a product; the bot transfers the customer to a human agent and a handoff notification fires.
- Lead capture — a separate extraction call records name, mobile, city, car brand, car model, and the product of interest, with a status (Interested / Recommended / Human Handoff).
Key Features¶
- Compatibility by car model — the distinctive requirement: model-aware product recommendation plus upselling, driven by editable fitment rules.
- Knowledge-base grounding — warranty, shipping, COD and FAQ answers come only from the configurable KB (uploadable PDFs/text in the full build).
- Rule-based human handoff — premium brands and complex cases route to a human with a notification.
- Lead capture and dashboard — leads plus an editable view of products, compatibility rules, and the knowledge base; the bot picks up changes without code.
- Channel-agnostic — connects to the client's WhatsApp Business number (Cloud API, QR connect).
Technology Stack¶
| Layer | Technology |
|---|---|
| LLM | Gemini 2.5 Flash via Vertex AI |
| Grounding | Catalogue + compatibility rules + knowledge base in the system context |
| Routing | Rule-based human-handoff brand list |
| Lead extraction | Separate JSON-mode call with schema and merge |
| Frontend / API | Next.js 15 (App Router), TypeScript, Tailwind CSS (WhatsApp-styled UI) |
| CRM store | File-backed JSON (swappable for the client's CRM) |
| Deployment | pm2 + nginx, Let's Encrypt SSL on a VPS |
Skills Demonstrated¶
- Domain-specific conversational AI (product recommendation with compatibility rules)
- Hybrid rule-based plus LLM routing (deterministic human handoff)
- Knowledge-base grounding and lead capture
- WhatsApp Cloud API chatbot architecture
- Vertex AI integration via a GCP service account