This sample provides a complete end-to-end solution for building, evaluating, and deploying a Retrieval Augmented Generation (RAG) retail copilot using Azure AI Foundry and Prompty. It's designed for developers and researchers looking to create AI-powered customer service applications grounded in specific business data, offering rapid prototyping and automated evaluation capabilities.

How It Works

The solution leverages a RAG architecture, utilizing Azure OpenAI for chat and embeddings, Azure AI Search for efficient data retrieval, and Azure CosmosDB for customer data storage. Prompty is integrated for prompt management and AI-assisted evaluation, enabling rapid iteration on prompt design and automated quality assessment of chatbot responses. The application is packaged as a FastAPI endpoint and deployed to Azure Container Apps, with infrastructure and deployment managed via the Azure Developer CLI (azd).

Quick Start & Requirements

• Install/Run: Use azd init -t contoso-chat to initialize, followed by azd up for provisioning and deployment.
• Prerequisites: Active Azure subscription, GitHub account, access to Azure OpenAI Services (with quota for gpt-4o-mini, gpt-4, text-embedding-ada-002), Azure AI Search with Semantic Ranker. Recommended Azure regions: eastus2 or francecentral. Familiarity with VS Code, Azure CLI, and Python is beneficial.
• Setup: GitHub Codespaces offers the fastest setup. Local setup requires Azure Developer CLI, Python 3.10+, Docker Desktop, and Git.
• Links: Prompty Documentation, Azure AI Foundry Documentation, Azure AI Templates.

Highlighted Details

• End-to-end GenAIOps workflow for RAG applications.
• AI-assisted evaluation with custom evaluators using Prompty.
• Deployment automation via Azure Developer CLI (azd) and GitHub Actions.
• FastAPI application deployed to Azure Container Apps.
• Integrates Azure OpenAI, Azure AI Search, and Azure CosmosDB.

Maintenance & Community

This is a Microsoft Azure sample. For community support, refer to the Code of Conduct and contact opencode@microsoft.com.

Licensing & Compatibility

The sample code is provided for demonstration purposes. It is advised not to use it directly in production without additional security features. The specific license for the sample code itself is not explicitly stated in the README, but it is a Microsoft Azure sample.

Limitations & Caveats

Some features used in this repository are in preview and not recommended for production workloads. The sample is intended for showcasing Azure services and tools, and requires careful security review before production deployment. Model availability and quota are region-dependent.