This project provides an AI-powered call center solution using Azure and OpenAI GPT models, targeting businesses needing automated customer service, IT support, or insurance claims handling. It enables AI agents to initiate and receive phone calls, offering real-time conversation streaming, multi-language support, and customizable voices, aiming to enhance customer experience and operational efficiency.

How It Works

The solution leverages Azure Communication Services for telephony and OpenAI's GPT-4o models for natural language understanding and generation. It employs Retrieval-Augmented Generation (RAG) with Azure AI Search for secure access to private data and follows best practices for handling sensitive information. Conversations are streamed in real-time, with features like call recording, human agent fallback, and customizable prompts for tailored interactions. Redis caching is used for performance optimization.

Quick Start & Requirements

• Installation: Deploy via Azure CLI using make deploy name=my-rg-name. GitHub Codespaces or local setup with Rust, uv, and Azure CLI is also supported.
• Prerequisites: Azure account with Communication Services, Cognitive Services, and OpenAI resources. A purchased phone number with voice capabilities is required.
• Resources: Estimated monthly cost for 1000 calls (10 mins each) is around $720.07, with detailed breakdowns for Azure services.
• Documentation: Demo available on YouTube; detailed deployment and customization guides are in the README.

Highlighted Details

• Supports GPT-4o and GPT-4o Realtime for advanced comprehension.
• Enables custom voice creation for brand consistency.
• Features real-time conversation streaming and resumption after disconnections.
• Includes robust data management with RAG, Redis caching, and structured claim schemas.

Maintenance & Community

The project is presented as a proof of concept, with a focus on demonstrating the integration of Azure services and OpenAI. The README indicates potential future enhancements and encourages contributions via pull requests.

Licensing & Compatibility

The project's licensing is not explicitly stated in the README. It is built on Azure services and OpenAI, which have their own terms of use. Compatibility for commercial use would depend on the underlying Azure and OpenAI service agreements.

Limitations & Caveats

This project is explicitly labeled as a proof of concept and not intended for production use without significant enhancements. Key areas for production readiness include comprehensive unit/integration tests, proper dashboarding, decoupling insights into a separate service, and implementing multi-region deployment and private networking for resiliency and security.