Azure ML

Azure Machine Learning is Microsoft's end-to-end ML platform for building, training, and deploying models within the Azure cloud ecosystem. It is the primary path for enterprises already invested in Microsoft tools to operationalize AI with enterprise-grade security, compliance, and governance.

• Managed compute — spin up GPU clusters for training and resize them automatically based on job queue depth
• Designer — build ML pipelines visually with a drag-and-drop interface, or define them in Python with the SDK
• Automated ML — automatically select algorithms, tune hyperparameters, and generate model explanations for tabular data
• Model catalog — deploy open-source models (Llama, Mistral, Phi) or Azure OpenAI models as managed inference endpoints
• Responsible AI dashboard — analyze model fairness, interpretability, and error patterns before deployment
• MLOps with Azure DevOps — build CI/CD pipelines for model retraining and deployment with Git-based version control

ML engineers and data scientists in Microsoft-stack organizations use Azure ML because it integrates natively with Azure Blob Storage, Azure Databricks, Synapse Analytics, and Power BI. It supports managed endpoints with blue/green deployment, making production rollouts safe and reversible.

Azure ML also offers Prompt Flow for building and evaluating LLM applications, providing a visual authoring experience for RAG pipelines and AI agents that connects directly to Azure OpenAI Service.

1. Create a workspace — in the Azure Portal, search for 'Machine Learning' and create a new workspace. You need an Azure subscription, a resource group, and a storage account.
2. Install the SDK — install the Azure ML Python SDK v2:

   pip install azure-ai-ml azure-identity

3. Authenticate — log in via the Azure CLI:

   az login
   az account set --subscription YOUR_SUBSCRIPTION_ID

4. Connect to your workspace — initialize the ML client in Python:

   from azure.ai.ml import MLClient
   from azure.identity import DefaultAzureCredential
   
   ml_client = MLClient(
       DefaultAzureCredential(),
       subscription_id='...',
       resource_group_name='...',
       workspace_name='...'
   )

5. Submit a training job — define a command job with your training script, environment, and compute target, then submit with ml_client.jobs.create_or_update().
6. Deploy a model — register the trained model and create a managed online endpoint with traffic routing and autoscaling rules.

Pricing: Pay for compute (VMs), storage, and endpoint hosting separately. An NC6s_v3 (V100 GPU) costs ~$3.06/hr. Free tier includes $200 credits for new Azure accounts. Full pricing details.