I built a WhatsApp sales assistant that answers property questions from a real brochure, and says “I don’t know” instead of making things up. Buyers message real estate teams at all hours with the same questions: What’s the price? Is it ready to move in? What amenities does it have? Someone has to answer each one, or the lead moves on. So I built a WhatsApp agent in n8n that answers them from the company’s own property catalogue, using RAG (retrieval-augmented generation). How it works: 📄 The property brochure PDF is downloaded, split into chunks and embedded into a vector store, so the agent can search the actual catalogue 🔍 Every product question goes through a retrieval step first. The agent only answers from what the catalogue says, not from what the model “thinks” it knows 🎙️ Customers can send voice notes. They get transcribed automatically, and the prompt handles Hyderabad locality spellings (Gachibowli, Kondapur, Narsingi, Madhapur and so on), because transcription loves to mangle those 🧠 Each customer gets their own chat memory, so follow-up questions like “and what’s the price for that one?” work naturally 🙋 If the catalogue has no answer, the bot says so honestly and points the customer to a human agent The goal isn’t to push a sale, it’s to give accurate, quick answers so the customer stays warm until a human steps in. Stack: n8n, WhatsApp Cloud API, Gemini, Ollama embeddings (nomic-embed-text), vector store retrieval. I build automations for real estate teams: lead qualification, FAQ bots and viewing scheduling. If your team is answering the same questions all day, message me. #n8n #RAG #AIAgents #RealEstate #WhatsAppAutomation #Automation #ChatbotDevelopment