Architecture

Customer calls (phone) → Twilio → Your server

OR

Customer uses web voice widget → Web Speech API → Your server

                 ↓
      Identify which company (tenant)
                 ↓
  Send customer’s speech to AI (OpenAI)
                 ↓
    ┌────────────┴────────────┐
    │                         │

Simple problem? Complex problem? │ │ ↓ ↓ AI answers directly AI creates a ticket │ │ ↓ ↓ AI speaks reply Route to available human agent │ │ ↓ ↓ End of call Agent takes over voice │ ↓ Call ends, transcript saved

The Flow

Absolutely — let me explain the flow of your AI customer support platform in plain English, like you’re describing it to a non-technical friend.

🧭 The Big Picture (30‑second version)

A customer calls your support number or speaks into a widget on your website. An AI voice agent answers immediately, listens to their problem, and tries to solve it. If the problem is simple (e.g., “Where’s my order?”), the AI handles it completely. If it’s complex or the customer is angry, the AI creates a ticket and routes the call to a human agent. The human sees all the chat history and can take over instantly.

Meanwhile, the business owner (tenant) sees everything that happens inside their own “separate bubble” – they never see other companies’ data.

🔁 Step‑by‑step flow (layman level)

1. The customer starts a conversation

• Phone call → they dial a real phone number (provided by Twilio).
• Web voice widget → they click a “Talk to Support” button on the company’s website and speak into their microphone.

2. The AI voice agent picks up

• The system listens to the customer’s words (using a speech‑to‑text engine).
• That text is sent to an AI brain (like GPT‑4) that belongs exclusively to that company (multi‑tenant isolation).

3. AI decides: can it solve the problem alone?