AIFondry-ContentUnderstanding

🏗️ Infrastructure - Deploy Content Understanding on Azure

Overview

This folder contains Infrastructure as Code for provisioning the Azure resource used by this repo: an Azure AI Services / AI Foundry account configured for Content Understanding.

The templates deploy:

• An Azure AI Services account with kind = "AIServices"
• SKU S0
• A custom subdomain so the Content Understanding endpoint is account-specific
• A system-assigned managed identity
• Public network access enabled
• Local authentication enabled so you can use API keys with the demos and quickstart

Notes:

• The Bicep deployment targets an existing resource group.
• The Terraform deployment creates both the resource group and the AI Services account.
• These templates provision the resource, but they do not create the required model deployments automatically.

Prerequisites

Before deploying, make sure you have:

1. An Azure subscription
2. The Azure CLI installed and authenticated with az login
3. One of the following:
   • Bicep CLI (or Azure CLI with Bicep support)
   • Terraform

Option A: Deploy with Bicep

The Bicep files are in infra\bicep.

1. Open infra\bicep\main.bicepparam
2. Update these values:
   • resourceName = globally unique Azure AI Services resource name
   • location = supported Content Understanding region
   • Optionally set customSubDomainName if you do not want it to match resourceName
3. Make sure your target resource group already exists
4. Run the deployment:

az deployment group create --resource-group <resource-group-name> --template-file .\infra\bicep\main.bicep --parameters .\infra\bicep\main.bicepparam

After deployment, note the endpoint output. It will be in this format:

https://<custom-subdomain>.cognitiveservices.azure.com

Option B: Deploy with Terraform

The Terraform files are in infra\terraform.

1. Open a terminal at the repo root
2. Change into the Terraform folder:

cd .\infra\terraform

1. Initialize Terraform:

terraform init

1. Preview the deployment:

terraform plan -var="resource_name=my-foundry-resource" -var="resource_group_name=rg-content-understanding" -var="location=eastus"

1. Apply the deployment:

terraform apply -var="resource_name=my-foundry-resource" -var="resource_group_name=rg-content-understanding" -var="location=eastus"

Terraform outputs include:

• endpoint
• resource_id
• resource_name

Note: variables.tf currently validates location against eastus, eastus2, westus, westus3, and westeurope. If you want to use another supported Content Understanding region, update that validation block first.

Post-Deployment Setup

1. Get the endpoint and key from the Azure Portal, or use Azure CLI:

az cognitiveservices account show --resource-group <resource-group-name> --name <resource-name> --query properties.endpoint -o tsv
az cognitiveservices account keys list --resource-group <resource-group-name> --name <resource-name>

1. Configure .env in the repo root. You can start from .env.template:

copy .env.template .env

Then set:

CONTENTUNDERSTANDING_ENDPOINT=https://<your-resource-name>.cognitiveservices.azure.com
CONTENTUNDERSTANDING_KEY=<your-api-key>

1. Enable model deployments in Content Understanding Settings:
   • https://contentunderstanding.ai.azure.com/settings
   • Add your deployed AI Foundry / AI Services resource
   • Enable the required models
2. Required models:
   • gpt-4.1
   • gpt-4.1-mini
   • text-embedding-3-large

Supported Regions

Content Understanding supports these regions:

• eastus
• eastus2
• westus
• westus3
• westeurope
• northeurope
• swedencentral
• uksouth
• southcentralus
• southeastasia
• australiaeast
• japaneast

Resources

• ARM template reference: Microsoft.CognitiveServices/accounts
• Content Understanding overview
• Language and region support

See Also

• Main README - Project overview, prerequisites, and full documentation links
• Quickstart Demo - Run your first demo in 2 minutes
• Demo Walkthrough Scripts - Progressive 3-step walkthrough
• Model Explanation - Why each AI model is needed

This site is open source. Improve this page.