All projects
Case study

AI-Powered WhatsApp Customer Support System

An AI-driven WhatsApp support system built for House & Hedges Real Estate to deliver instant, accurate 24/7 replies to prospective clients — orchestrated with n8n, backed by PostgreSQL and Evolution API, and grounded in the company's own policies and knowledge base.

4 phases8 technologies
AI-Powered WhatsApp Customer Support System
  1. 01
    Phase 1: Business Problem & System Design

    Designing 24/7 Instant, Accurate Customer Support for a Real Estate Firm

    House & Hedges Real Estate needed a way to answer prospective clients around the clock with information that was both instant and accurate — even outside office hours. Instead of hiring a night shift or paying for a third-party chatbot subscription, I designed a self-hosted AI support system on top of WhatsApp, the channel our audience already uses. The architecture was intentionally decoupled: n8n for orchestration, PostgreSQL for state and history, Evolution API as the WhatsApp bridge, and a large language model to generate on-brand responses.

  2. 02
    Phase 2: Local Dockerized Development Environment

    Containerizing n8n, PostgreSQL and Evolution API for Rapid Iteration

    I began development on my local machine by containerizing the entire stack with Docker. n8n hosted the automation flows, PostgreSQL persisted the WhatsApp instances, conversations and chat history, and Evolution API handled the WhatsApp platform integration. Running everything as isolated containers made the setup portable and let me iterate on the automation without touching the host system.

  3. 03
    Phase 3: Automation Flow, Webhooks & AI Response Generation

    Wiring WhatsApp → Evolution API → n8n → PostgreSQL → LLM → Reply

    After connecting a WhatsApp number to Evolution API, I registered n8n as its webhook target so every incoming message triggered an automation run. Inside n8n the flow logged the message to PostgreSQL, filtered by message type and intent to decide the next action, and assembled a context payload from the ongoing chat history together with the company's terms, policies and property information. That context was sent to an LLM to generate an on-brand reply, which was then dispatched back to the customer through Evolution API's send-message webhook — closing the loop in seconds.

  4. 04
    Phase 4: Production Deployment on Company VPS

    Migrating to the Company VPS and Scaling to Multiple WhatsApp Numbers

    Once the automation was validated end-to-end locally, I moved the entire Dockerized stack to the company's VPS for 24/7 operation. The final production system supports multiple WhatsApp numbers with a separate interface per number, keeps per-number chat records, and delivers informative, accurate responses grounded in the firm's own knowledge — giving House & Hedges an always-on support channel without recurring per-message SaaS fees.