Built an AI Real Estate Agent for Dubai Brokers (WhatsApp + n8n + OpenAI)

Built with n8n
, , , ,
1

Hi n8n community! :waving_hand:

I wanted to share a recent automation architecture I built for the Dubai real
estate market.

The Problem: In Dubai’s highly competitive market, real estate brokers lose hot
leads because they can’t reply to WhatsApp messages instantly 24/7. Buyers
expect immediate answers about property availability, complex payment plans, and
brochures.

The Solution: I built an autonomous AI assistant powered by n8n that handles the
entire preliminary sales cycle, qualifies the lead, and collects booking
documents—completely on autopilot.

Architecture & Workflow: Here is a high-level overview of the pipeline:

  1. Trigger: Incoming WhatsApp messages via Webhook (using Whapi.cloud API).
  2. Context & RAG: Google Sheets nodes dynamically fetch the anti-spam
    stop-list, property availability, and FAQs.
  3. AI Brain: OpenAI Chat Model (GPT-4o) combined with the AI Agent node
    processes the user’s intent. Custom JS code structures the RAG data.
  4. Execution: The workflow uses a Switch/If logic to determine if it should
    send a standard text reply or dynamically attach a PDF brochure (via HTTP
    Request) based on the client’s request.

AI Assistant
AI Assistant1920Ă—869 99.9 KB

Key Features of the Agent:

  • Dynamic Content Delivery: Instantly shares Google Drive links or PDF
    brochures based on user requests (e.g., “Looking for a 2-bedroom”).
  • Objection Handling: Explains complex Payment Plans (e.g., 20/80, 40/60) and
    Handover dates using strictly the data provided in the Sheets (Zero
    hallucinations).
  • Intent Recognition: Automatically requests KYC documents (Passport, Emirates
    ID, Proof of Payment) when the client triggers a “ready to book” intent.

Demo Video: You can see a quick 1-minute demo of how the AI interacts with a
client in real-time here:

Template: I have attached the sanitized structural JSON workflow below for
anyone who wants to see the node setup. (Note: Client credentials, specific
Google Sheet IDs, and advanced JS parsing logic have been removed for privacy).

I specialize in building complex, production-ready AI architectures and data
pipelines. If anyone here is building something similar, has questions about the
architecture, or wants to collaborate on B2B projects, let’s connect on
LinkedIn! :link:

AI_Assistant_clean.json (12,4 КБ)

Happy automating! :rocket:

4 Likes
2

Welcome @Dmitry_ai_automation!

Using Google Sheets as a live RAG layer for property availability and FAQs is a practical choice - it stays editable without redeployment, which brokers love. One thing worth planning for as the listing count grows: fetching and scanning all rows per conversation gets expensive in latency and API calls. When you hit ~200+ properties, a small Qdrant (or Pinecone) vector store with property embeddings plus a semantic search step before the AI agent will cut context size and keep responses faster. The anti-spam stop-list approach via Sheets is a clean touch.

1 Like
3

Hi Jay, thanks for the warm welcome and the insightful feedback! :raising_hands:

You are 100% spot on regarding the latency and API cost at scale. Using Google Sheets was a deliberate choice for Phase 1 (MVP) purely for the client’s operational comfort—brokers love having a familiar interface where they can manually tweak prices or availability on the fly without needing a CMS dashboard.

But you are absolutely right: once the agency scales past ~150-200 active listings, migrating that RAG layer to Pinecone or Qdrant is the mandatory next step to keep the AI responses snappy and context windows optimized.

Glad you liked the stop-list approach! Cheers.

1 Like