⚠️
This project is currently in active development and may contain breaking changes.
Updates and modifications are being made frequently, which may impact stability or functionality. This notice will be removed once development is complete and the project reaches a stable release.
This project implements the Azure AI Baseline Reference Architecture to deploy an Azure AI Foundry following best practices for networking, security, and model integration. It provisions the infrastructure necessary for chatting over data using managed models such as:
- Ada-002 for text embeddings
- GPT-4o for natural language generation
- Phi-4 open mulit-model deployed as serverless api
The solution automates the deployment of Azure AI Foundry, including AI Hubs, AI Projects, and networking components, using PowerShell and Bicep templates. It incorporates observability through Log Analytics and Application Insights for real-time monitoring and diagnostics. Additionally, it integrates Azure Functions with Blob Triggers for event-driven processing of JSON documents, chunking document text, extracing meta data, generating vector embeddings with the Ada-002 model, and indexing into Azure AI Search to enable retrieval-augmented generation (RAG) and semantic search that will be tested using the PlayGround.
-
Azure AI Foundry Deployment:
- Deploys AI Hub, AI Project, and AI Services for model hosting with secure access via private endpoints and managed identities.
-
Networking and Security:
- Creates a VNet with subnets, a VPN Gateway for remote access, and private endpoints for secure communication.
-
Model Integration:
- Enables chat over data with GPT-4o and Phi-4.
- Supports vector search and RAG with Azure AI Search.
- Generates Ada-002 vector embeddings from JSON documents.
-
Vector Processing Pipeline:
- Uses Azure Blob Storage triggers to process new JSON files with Azure Functions.
- Extracts content, generates Ada-002 embeddings, and indexes them in Azure AI Search for semantic retrieval and RAG.
-
Infrastructure as Code:
- Automates deployment with Bicep and PowerShell using modular, reusable templates.
Azure AI Foundry Reference Architecture
This project builds on the Azure AI Baseline Reference Architecture to help you design and deploy enterprise-grade generative AI solutions. It incorporates networking, security, and authorization best practices, enabling a scalable and secure AI environment.
🔗 Azure AI Baseline Reference Architecture
🔗 Azure AI Foundry Agents Network-Secured Environment
🔗 Private storage configuration
Please review the quickstart templates that demonstrates how to set up Azure AI Foundry with a network-restricted configuration.
🔗 Azure AI Foundry Template - Network Restricted
🔗 Azure AI Foundry Agents Template - Network Restricted
AI Model Deployment in Azure AI Foundry
To explore model deployment options, including serverless models, fine-tuning, and inference endpoints, refer to the official documentation.
🔗 Deploy AI Models in Azure AI Foundry Portal
🔗 Deploy models as serverless APIs
Follow these key steps to successfully deploy Azure AI Foundry:
- Detailed instructions for deploying Azure AI Foundry, including prerequisites, configuration steps, and setup validation.
- Step-by-step guide for processing and indexing documents into Azure AI Search to enable vector search within Azure AI Foundry.
- Instructions for building a QnA agent using Azure AI Agent Services, powered by Azure AI Search as the vector-based knowledge store.
- How to test and verify the online endpoint hosting a Hugging Face model to ensure successful deployment and connectivity.
After completing testing, ensure to delete any unused Azure resources or remove the entire Resource Group to avoid incurring additional charges.
This project is licensed under the MIT License, granting permission for commercial and non-commercial use with proper attribution.
This demo application is intended solely for educational and demonstration purposes. It is provided "as-is" without any warranties, and users assume all responsibility for its use.