This repository provides a Python-based solution accelerator for the Retrieval Augmented Generation (RAG) pattern, targeting developers who need to customize beyond Azure OpenAI's out-of-the-box capabilities. It enables natural language querying of private data using Azure OpenAI and Azure AI Search, offering features like document upload, web page indexing, and speech-to-text.

How It Works

The accelerator implements a RAG pattern using Azure OpenAI for LLM capabilities and Azure AI Search for retrieval. It supports various data ingestion methods (push/pull) and orchestration frameworks (Semantic Kernel, LangChain, Prompt Flow). Users can configure chunking strategies, prompts, and retrieval types, with options for PostgreSQL or Cosmos DB for data storage and chat history.

Quick Start & Requirements

• Install/Run: One-click "Deploy to Azure" button or local deployment instructions.
• Prerequisites: Azure subscription with owner access, approval for Azure OpenAI services. Optional: Microsoft 365 for Teams extension.
• Resources: Deploys Azure App Service, Application Insights, Bot Service, OpenAI, Document Intelligence, Functions, AI Search, Storage, Speech, CosmosDB/PostgreSQL.
• Docs: Supporting documentation

Highlighted Details

• Supports grounding models with both private data and public web pages.
• Offers an admin site for dataset ingestion, inspection, and configuration.
• Provides flexibility in choosing between PostgreSQL and Cosmos DB for data management.
• Includes speech-to-text functionality for natural language interaction.

Maintenance & Community

This is an Azure sample, maintained by Microsoft. Further community engagement details are not explicitly provided in the README.

Licensing & Compatibility

• License: MIT License for the repository. CDLA-Permissive-2 License for sample data.
• Compatibility: Designed for Azure services. Use of third-party components requires compliance with their respective licenses. Not intended for medical or financial advice.

Limitations & Caveats

The solution is a baseline sample and not intended for production use without evaluation and customization. Specific Azure OpenAI model versions may have regional availability constraints. The solution is not subject to SOC 1 and SOC 2 compliance audits.