AzureML_industry_labs: Instructions file for Codex

AGENTS.md

This document describes the workspace structure, conventions, and patterns for AI coding agents working in the Azure ML Industry Labs repository. Use this as context when generating, modifying, or reviewing labs.

Overview

Azure ML Industry Labs (azureml_industry_labs) is a collection of end-to-end machine learning labs across different industry use-cases, built on Azure Machine Learning. Each lab demonstrates a complete MLOps workflow — data preprocessing, model training, model registration, and deployment — using the Azure ML SDK v2, MLflow experiment tracking, and managed endpoints.

Labs are self-contained Python projects (not notebooks) with a shared structural convention.

---

Directory Structure

Root

azureml_industry_labs/
├── README.md              # Repo overview, labs table, quick start
├── CONTRIBUTING.md         # Guidelines for adding new labs
├── AGENTS.md               # This file — workspace context for AI agents
├── LICENSE                 # MIT License
├── images/                 # Shared images and diagrams
│
└── <lab_name>/             # Each lab is a top-level directory

Lab Structure (Standard Template)

Every lab must follow this structure. When creating a new lab, replicate this layout exactly:

<lab_name>/
├── main.py                  # Pipeline orchestration & job submission
├── lab.json                 # Lab metadata (used by CI to update root README)
├── requirements.txt         # Python dependencies (pinned versions)
├── Dockerfile               # Custom Azure ML environment image
├── README.md                # Lab-specific documentation
├── .amlignore               # Files to exclude from Azure ML snapshots
├── .gitignore               # Files to exclude from version control
├── data_processing/
│   ├── __init__.py
│   └── preprocess.py        # Reusable dataset class (PyTorch Dataset)
├── model/
│   ├── __init__.py
│   └── <model_name>.py      # Model architecture definition
└── pipeline/
    ├── preprocess_step.py   # Data preprocessing pipeline step
    ├── train_step.py        # Model training pipeline step
    ├── register_model.py    # Model registration pipeline step
    ├── deploy_endpoint.py   # Endpoint deployment pipeline step
    └── score.py             # Scoring script for batch/online inference