WhatsApp RAG Bot: E-commerce support with Qdrant
WhatsApp AI agent for an electronics store with a vector knowledge base (Qdrant). Auto-indexing of Google Drive documents, conversational memory, and a RAG system to respond accurately about products, technical support and after-sales service.
Knowledge base
Response
Memory
Context
An electronics store gets dozens of identical WhatsApp questions every day: product differences, troubleshooting, order status, returns. The support team spends its time copy-pasting the same answers from the product docs. Meanwhile, customers wait and satisfaction drops.
The challenge
Build a WhatsApp agent that answers accurately by relying on the official product docs (no hallucinations) while keeping the conversation fluid. The docs evolve regularly, the bot must update automatically whenever a new file lands in the team's Google Drive.
The solution
4-step n8n pipeline. (1) Qdrant collection creation. (2) Indexing: Google Drive → Download Files → Token Splitter (chunk 300, overlap 30) → OpenAI Embeddings → Qdrant Vector Store. (3) Meta webhook verification (GET hub.challenge + POST messages). (4) Conversational AI Agent gpt-4o-mini with Window Buffer Memory + RAG tool (toolVectorStore 'company_data' connected to Qdrant), strict electronics-support system prompt, then WhatsApp send.
Key features
- 01Automatic Qdrant collection creation + refresh
- 02Google Drive indexing: any new file becomes instant knowledge
- 03Optimized Token Splitter (chunk 300, overlap 30) for RAG precision
- 04OpenAI embeddings for vectorization
- 05Meta webhook verification (GET hub.challenge) + message reception (POST)
- 06Text-only filtering (ignores WhatsApp status notifications)
- 07Fallback response on non-text: 'You can only send text messages'
- 08Conversational AI Agent gpt-4o-mini with electronics-support system prompt
- 09Window Buffer Memory for conversation tracking
- 10RAG tool 'company_data' retrieves info from Qdrant at every turn
- 11Final response sent via WhatsApp Cloud API