Azure-AI_Foundry

Lesson 2: Project Management

Organize your AI work within projects — the development workspace inside a hub.

What is a Project?

A project is a workspace within a hub where you develop AI applications. Each project has its own:

• Model deployments and endpoints
• Data assets and connections
• Evaluation results and experiments
• RBAC permissions (who can access what)

📖 Docs: Create a Project

---

Creating a Project

Portal

1. Open ai.azure.com → select your hub
2. Click + New project
3. Name the project and configure settings

Azure CLI

az ml workspace create \
  --name my-project \
  --resource-group rg-ai-foundry \
  --kind project \
  --hub-id /subscriptions/<sub>/resourceGroups/<rg>/providers/Microsoft.MachineLearningServices/workspaces/<hub>

Python SDK

from azure.ai.ml.entities import Project

project = Project(
    name="my-project",
    hub_id="<hub-resource-id>"
)
ml_client.workspaces.begin_create_or_update(project).result()

---

Project-Level RBAC

Azure AI Foundry uses Azure RBAC to control access. Key roles:

Role	Permissions
Azure AI Developer	Full project access — deploy models, run evaluations, manage data
Azure AI Inference Deployment Operator	Deploy and manage endpoints only
Reader	View-only access to project resources
Owner	Full control including RBAC management

# Assign AI Developer role to a user for a specific project
az role assignment create \
  --assignee user@contoso.com \
  --role "Azure AI Developer" \
  --scope /subscriptions/<sub>/resourceGroups/<rg>/providers/Microsoft.MachineLearningServices/workspaces/<project>

💡 Tip: Follow least-privilege principles — give developers the Azure AI Developer role, not Owner. Reserve Owner for platform admins.

📖 Docs: RBAC for AI Foundry

---

Project Assets

Each project contains:

Asset	Description
Models	Deployed model endpoints (OpenAI, serverless, managed)
Data	Uploaded datasets, documents for RAG
Evaluations	Evaluation runs and results
Indexes	AI Search indexes connected to the project
Flows	Prompt Flow definitions
Endpoints	Deployed model and flow endpoints
Compute	Shared from hub or project-specific

---

Sharing Models Across Projects

Models deployed at the hub level can be shared across multiple projects. This is useful when:

• Multiple teams need access to the same GPT-4o deployment
• You want to centralize model governance and cost tracking
• Consistent model versions across dev/staging/prod

To share a connection:

1. Create the Azure OpenAI deployment in the hub
2. All projects within that hub automatically inherit the connection

---

Project Quotas and Limits

Resource	Limit
Projects per hub	100
Connections per project	50
Deployments per project	Varies by model
Data assets per project	500

💡 Tip: Use a naming convention like {team}-{app}-{env} for projects (e.g., platform-chatbot-dev). This makes it easy to find and manage projects at scale.

---

Best Practices

1. One project per application per environment — myapp-dev, myapp-staging, myapp-prod
2. Use project-level RBAC — don’t give everyone hub-level access
3. Tag projects with team, application, cost-center
4. Centralize shared models at hub level — deploy once, use everywhere
5. Review quotas regularly — set up alerts before hitting limits