Learning Timeline
Key Insights
Data Compliance Tips (GDPR)
If your company operates in Europe or needs to comply with GDPR, select Azure as your transcription provider to ensure data remains within the European continent.
Controlling Agent Behavior
The AI Assistant will strictly follow the 'System Prompt'. If you configure it solely for scheduling, it may reject other requests (such as language translation) even if the base model (GPT-4o) is capable of performing them.
The Advantages of Generative AI in Voice
Voice transcription accuracy has increased fivefold since 2012. Using models like GPT-4o enables much more natural conversations compared to traditional bank IVR systems.
Prompts
Scheduling Assistant System Prompt Example
Target:
Vapi Assistant System Prompt
You are Riley, a scheduling assistant for Wellness Partners. Your primary goal is to assist users with scheduling appointments. Be professional, clear, and helpful. Do not perform tasks outside of scheduling, such as translation or general knowledge queries, unless they relate to the appointment.
Step by Step
How to Build an AI Voice Assistant in Vapi
- Log in to the Vapi platform and go to the main dashboard.
- Click on the 'Assistants' menu in the sidebar to view your list of voice agents.
- Click the 'Create Assistant' button or select an existing Assistant to start configuration.
- In the 'Model' section, choose the AI that will serve as the 'brain' of your agent (e.g., GPT-4o or Claude 3.7 Sonnet).
- Select a 'Transcribe' provider to convert user speech to text (choose Deepgram for speed or Azure if you require GDPR compliance).
- Choose a 'Voice' provider to define the AI’s synthesized voice (e.g., voices from GPT-4o for more authentic quality).
- Fill in the 'System Prompt' in the designated field to set the agent's personality and task scope.
- Configure 'Actions' if you want the AI to perform external tasks like sending texts or updating Google Sheets.
- Click the 'Talk to Assistant' button at the bottom right of the screen.
- Click 'Allow' when the browser requests microphone access to start talking to your AI.
