Build a Complete Voice AI System with Claude Code (Step-by-Step)

Voice AI System Development with Claude Code

Key Concepts:

• Claude Code: An AI-powered coding assistant used to automate the creation of voice AI systems.
• Retail AI: A voice agent platform used for building and deploying conversational AI applications.
• Naden: A workflow automation platform used for connecting applications and automating tasks.
• Agentic Workflows: Automated sequences of actions performed by AI agents.
• Webhooks: Automated messages sent from one application to another when a specific event occurs.
• JSON Payload: A data format used for transmitting information between applications.
• API Key: A unique identifier used to authenticate and authorize access to an application's API.
• Prompt Engineering: The process of crafting effective prompts to guide AI models.
• Production vs. Test Environments: Distinguishing between live, customer-facing systems and development/testing environments.

---

I. Introduction & System Overview

The video demonstrates the creation of a complete voice AI system for John’s HVAC company using Claude Code, minimizing manual setup and coding. The system, built entirely through prompting Claude, includes a voice-enabled front-end (via Retail AI), comprehensive functions, post-call data extraction, and three distinct workflows: calendar availability check, appointment booking, and post-call routing. The speaker emphasizes the transformative potential of this approach, highlighting the ability to build functional voice AI systems without a coding background. The completed system is demonstrably functional, with successful workflow executions indicated by green dots in the Naden interface.

II. Setting Up the Development Environment

The initial step involves installing the Claude Code desktop application (available at cloud.ai). Crucially, a Pro plan subscription ($20/month) is required to access the necessary features. Within Claude Code, users navigate to the “Co-work” section (instead of “Chat”) to utilize the AI-powered coding assistant. A dedicated folder on the user’s local computer is created to store the project files, ensuring local access and organization.

III. Prompting Strategies: Comprehensive vs. Step-by-Step

Two primary prompting methods are presented:

• Comprehensive Prompt: A single, detailed prompt containing all instructions for building the system. This is the method demonstrated in the video.
• Step-by-Step Prompt: A series of prompts, each focusing on a specific aspect of the system’s development (e.g., prompt 1 for initial setup, prompt 2 for function creation).

The speaker directs viewers to a free community resource (link in description – classroom section, YouTube resources, “Claude Code HVAC voice”) where both prompt types are available for download. A paid community (Retail AI certified program) offers a more in-depth certification program focused on building client-ready voice agents.

IV. The Comprehensive Prompt & System Requirements

The comprehensive prompt instructs Claude to build a voice scheduling system for an HVAC company ("Coolflow HVAC"). Key requirements outlined in the prompt include:

• Greeting & Information Gathering: The agent should greet callers, inquire about the issue (AC, heating, maintenance, etc.), and collect name, phone number, and service address.
• Urgency Assessment: The agent should determine if the issue is urgent.
• Function Calls via Naden: The system must utilize real function calls through Naden, not mock functions.
• Specific Functions: The prompt explicitly requests functions for checking availability, booking appointments, and post-call data handling.
• Integration with Google Calendar: Availability checking and appointment booking should integrate with Google Calendar.
• Webhook Endpoints: Exposing webhook endpoints for post-call actions.
• Post-Call Data Extraction: Collecting caller information (name, phone, address, issue).
• Automated Routing: Sending Slack messages for urgent issues and emails for appointment confirmations.

V. API Key Configuration & Integration

After pasting the prompt into Claude Code, the system requests several API keys and URLs:

• Retail AI API Key: Obtained from the Retail AI settings page (API Keys section).
• Naden Cloud Instance URL: Found on the Naden homepage (e.g., yourname.app.cloud).
• Naden API Key: Generated from the Naden settings page (Naden API section).
• Gmail Address: Used for sending appointment confirmation emails.
• Slack Webhook URL: (Optional) Used for sending urgent issue notifications.

The speaker provides detailed instructions on locating and obtaining each key, emphasizing the importance of keeping API keys secure.

VI. System Creation & Workflow Generation

Once the API keys are provided, Claude Code automatically creates the necessary components:

• Naden Workflow: Three workflows are generated: one for checking availability, one for booking appointments, and one for post-call webhooks. These workflows are automatically published and have production URLs.
• Google Calendar Integration: The system integrates with Google Calendar for availability checking and event creation. Users must manually connect their Google Calendar account within Naden.
• Retail AI Agent: A voice agent named "Coolflow HVAC" is created in Retail AI, complete with a system prompt and integrated functions.
• Webhook Connections: Webhooks are automatically configured to connect Retail AI to Naden workflows.
• Post-Call Data Extraction: The system extracts caller information and sends it to Naden.

VII. System Demonstration & Functionality

The speaker demonstrates the completed system, showcasing:

• Naden Workflow Details: The three generated workflows are visible in Naden, with clear connections to webhooks and Google Calendar.
• Retail AI Agent Configuration: The Retail AI agent’s system prompt and function integrations are displayed.
• JSON Payload: The JSON payload sent to Naden via webhooks is shown, demonstrating the data transfer process.
• Post-Call Data Extraction: The system’s ability to extract caller information is highlighted.

VIII. Limitations & Future Development

The speaker acknowledges that while Claude Code automates a significant portion of the development process (approximately 80%), some manual adjustments and refinements are still necessary, particularly for production-level deployments. The speaker announces plans to create more content focused on Claude Code within the paid community, including advanced prompting techniques and building more complex voice agent systems.

Notable Quote:

“This is absolutely mind-blowing how easy things have become.” – Speaker, referring to the automation capabilities of Claude Code.

Data/Statistics:

• Claude Code Pro plan cost: $20/month.
• Automation level achieved: Approximately 80% of the system development is automated.

Conclusion:

The video effectively demonstrates the potential of Claude Code to drastically simplify the development of voice AI systems. By leveraging AI-powered prompting, developers can automate the creation of complex workflows and integrations, significantly reducing the need for manual coding and setup. While some manual refinement is still required, the demonstrated system represents a significant step towards democratizing access to voice AI technology.