AI Voice Assistant – Interview Assignment

📌 Project Overview

Design a mini system for a dental marketing AI voice assistant that integrates with GoHighLevel (GHL).
The system should:

1. Receive new leads from GHL (via webhook).
2. Initiate an outbound call using Twilio (or another voice API).
3. Use AI (STT + LLM + TTS) to qualify the lead and collect structured info.
4. Update the lead’s status back in GHL (notes, stage, or appointment).

---

🎯 Objectives

• Show system architecture and data flow.
• Provide code snippets for key interactions.
• Demonstrate compliance awareness (DNC, PHI, escalation).

---

🛠 Requirements

1. Architecture

• Diagram or clear explanation of the flow:
  GHL Webhook → Backend → Voice Provider → AI Pipeline → GHL Update

2. Code Examples

Provide working or pseudocode for:

• Handling a GHL webhook (new lead).
• Initiating a Twilio outbound call.
• LLM prompt design to capture structured data like:

  {
    "pain_level": "7/10",
    "insurance": "Sun Life",
    "preferred_slot": "2025-09-07 15:30"
  }

• • Writing results back to GHL (appointment or note)

• 3. Compliance

When designing a system that handles patient calls, compliance and safety are critical. In your solution, explain how you would address:

• Do Not Disturb (DND) / Do Not Call (DNC)

  • Respect the flags from GHL to avoid calling leads who opted out.

• Protected Health Information (PHI)

  • Collect only minimal data necessary for appointment booking (chief complaint, pain scale, insurance provider).
  • Avoid storing sensitive information beyond what is needed.
  • Ensure all stored data is encrypted at rest and in transit.

• Escalation & Safety

  • Define red-flag conditions (e.g., severe pain 8–10, swelling + fever, trauma, allergies).
  • In such cases, the AI should immediately transfer to a human receptionist or dentist.
  • Clearly state that the AI cannot provide clinical advice.

---

📂 Deliverables

• A pull request to this GitHub repo containing:
  • Architecture diagram (flow from GHL → AI voice pipeline → GHL update).
  • Code snippets (Node.js, Python, or your preferred language).
  • Notes on compliance and design decisions.
  • (Optional) Example conversations/transcripts showing how the AI responds.

---

⚙️ Suggested Tech Stack

• Voice Layer: Twilio Programmable Voice (or SignalWire)
• Speech-to-Text (STT): Whisper API or Google Cloud Speech-to-Text
• Language Model (LLM): GPT-4o (or similar) for dialogue + tool calling
• Text-to-Speech (TTS): Amazon Polly or ElevenLabs
• Backend: Node.js (Express) or Python (FastAPI) for handling webhooks and call logic
• Database (optional): Postgres (to store call logs, transcripts, and metrics)

---

📅 Submission

• Submit your solution either as:

  1. A GitHub repo link, or
  2. A PDF/Markdown doc with architecture, code snippets, and explanations.

• Deadline: 3 days after receiving this assignment.

---

🔖 Reference Payloads

Example GHL Webhook (New Lead)

{
  "event": "contact.created",
  "contact": {
    "id": "CNT_xxx",
    "firstName": "Sarah",
    "phone": "+1416xxxxxxx",
    "email": "sarah@example.com"
  }
}
### Example Twilio Outbound Call (Node.js)
```js
import twilio from "twilio";

const client = twilio(
  process.env.TWILIO_ACCOUNT_SID,
  process.env.TWILIO_AUTH_TOKEN
);

async function makeCall(toNumber, greeting) {
  try {
    const call = await client.calls.create({
      to: toNumber,
      from: process.env.TWILIO_NUMBER,
      twiml: `<Response>
                <Say voice="alice">${greeting}</Say>
                <Pause length="5"/>
                <Say voice="alice">Thank you. Goodbye.</Say>
              </Response>`
    });

    console.log(`Call initiated. SID: ${call.sid}`);
    return call.sid;
  } catch (err) {
    console.error("Error initiating call:", err);
    throw err;
  }
}

// Example usage
makeCall("+1416xxxxxxx", "Hello Sarah, this is Nova Dentistry calling about your appointment request.");