MLOps Templates¶

The AI Factory provides MLOps templates for the ESML (Enterprise Scale Machine Learning) project type, enabling a full ML lifecycle within a secure, private Azure environment.

ESML Architecture¶

The ESML project type includes:

Azure Machine Learning workspace — private, with private endpoints on all dependent services.
AML Compute Clusters — auto-scaling, configurable SKU and node count per environment.
AML Compute Instances — for interactive development.
Private AKS Cluster — for model serving at scale (Azure Arc-enabled).
Azure Container Registry (ACR) — private, Premium SKU for model image storage.
Azure Data Factory — for orchestrating training and batch-inference pipelines.
Databricks (optional) — for large-scale feature engineering.

ESML With Fabric Flavour¶

The ESML project type optionally integrates with Microsoft Fabric:

Data scientists can work from Azure Databricks, Microsoft Fabric, or Azure Machine Learning — with the same MLOps template.
Fabric provides a unified analytics platform with OneLake as the data foundation.

MLOps Pipeline Templates¶

Templates are provided for:

Template	Description
Training pipeline	AML pipeline for model training with logging and experiment tracking
Batch inference pipeline	Scheduled or event-driven batch scoring
Online inference	AKS-hosted REST endpoint with private networking
Model registration & promotion	Automated model registration and promotion across Dev → Stage → Prod

Compute Defaults (Overridable)¶

Setting	DEV default	TEST/PROD default
AKS node SKU	`Standard_B4ms`	`Standard_DS13-2_v2`
AKS node count	1	3
AKS Kubernetes version	`1.33.2`	`1.33.2`
AML cluster max nodes	3	5
AML cluster SKU	`Standard_DS3_v2`	`Standard_D13_v2`
AML compute instance SKU	`Standard_DS11_v2`	`Standard_ND96amsr_A100_v4`

All compute settings are overridable via admin_aks_* and admin_aml_* variables.

Info

MLOps templates are located under copy_my_subfolders_to_my_grandparent/mlops/.