Hands-on Azure Machine Learning (DP-100) and Microsoft AI Foundry / Azure AI Studio (AI Engineer) notebooks, artifacts, and cheatsheets — covering classic ML, MLOps, Responsible AI, privacy techniques, and GenAI patterns (RAG + fine-tuning).
Sources & attribution: This repo is based on and extends the official MicrosoftLearning labs:
- Azure ML (DP-100): https://github.com/MicrosoftLearning/mslearn-azure-ml/blob/main/index.md
- AI Studio / AI Foundry: https://github.com/MicrosoftLearning/mslearn-ai-studio/blob/main/index.md
Content here is expanded with additional explanations, examples, screenshots, troubleshooting, and reusable templates.
NOTE: Repository in work/progress phase some section will be updated and reworked.
- What you’ll find here
- Repository layout
- Notebooks index
- Prerequisites
- Quickstart
- Responsible AI (RAI) dashboards
- Privacy & SmartNoise [in progress]
- GenAI: RAG + fine-tuning [in progress]
- Cheatsheets
- Artifacts & large files
- Contributing
- License & disclaimer
- Acknowledgements
This repository is designed as a single hub for practical, exam-aligned, real-world Azure ML + AI Foundry work:
- Training & evaluation notebooks (regression/classification)
- Custom environments and reproducible runs
- Hyperparameter tuning
- MLflow tracking + model registry patterns
- Pipelines (training → scoring → evaluation)
- Deployment endpoints (managed online endpoints / batch patterns)
- Model artifacts & run outputs (where appropriate)
- Responsible AI dashboards (model explanations, error analysis, fairness)
- Reproducible evaluation patterns and reporting
- Notes on common pitfalls & interpretation tips
- Privacy-aware data practices and examples
- Usage examples with SmartNoise (or similar DP tooling where applicable)
- RAG with Azure AI Search
- Promptflow / orchestration patterns (where used)
- Fine-tuning patterns for LLMs (where supported/available in your tenant)
- Practical evaluation notes for LLM apps (quality + safety)
- ML/LLM metrics cheat sheets + small examples
- Simplified neural network definitions (MLP/CNN/RNN/GAN, etc.)
- ML algorithms explained in “plain English”
- Troubleshooting playbook for common Azure issues
Located under: notebooks/dp100-azure-ml/
-
01-train-custom-environments-automl.ipynb
Custom environments, training runs, and AutoML patterns. -
02-diabetes-class-mlflow-tracking.ipynb
Experiment tracking with MLflow, logging metrics/artifacts. -
03-hyperparm.ipynb
Hyperparameter tuning patterns and sweep concepts. -
03-ml-flow.ipynb
MLflow fundamentals (tracking + model packaging patterns). -
04-pipelines.ipynb
Pipelines: training → scoring → evaluation. -
05-responsible-ai.ipynb
Responsible AI: explanations, error analysis, fairness workflow. -
06-mlflow-autolog.ipynb
Autologging patterns and how to interpret runs/metrics. -
07-endpoints.ipynb
Deployment endpoints and scoring patterns.
Located under: notebooks/ai-foundry/
-
rag-azure-ai-search/
Indexing, retrieval, grounding patterns, evaluation notes. -
fine-tuning/
Fine-tuning workflows (where supported), dataset formatting, evaluation. -
evaluation/
LLM app evaluation patterns (quality + safety), logging, dashboards.
You can run most notebooks directly in Azure ML Studio / AI Foundry.
For local execution, you'll typically need:
-
An Azure subscription with access to:
-
Azure ML workspace (DP-100 content)
-
AI Foundry / Azure AI Studio (GupenAI content)
-
Azure Storage (used by both)
-
(GenAI) Azure AI Search (for RAG)
-
-
Azure CLI + ML extension (optional but useful)
-
Python 3.10+ (recommended)
-
Permission to create compute / endpoints (or a pre-provisioned environment)
-
Create or open your Azure ML Workspace or AI Foundry Project.
-
Upload the relevant notebook folder from
notebooks/. -
Select a compute (or create one).
-
Run cells top-to-bottom.
-
Clone the repo:
-
az login
Notes:
Some notebooks assume Studio-managed identities and compute. Local execution may require extra auth setup.
For endpoint deployments, ensure your account has permission to create managed online endpoints.
Responsible AI dashboards typically include:
-
Model explanations (global/local feature importance)
-
Error analysis (where predictions fail)
-
Fairness analysis (distribution + disparity insights)
Recommended practice:
-
Keep evaluation datasets versioned (or reproducible).
-
Save key dashboard screenshots under:
-
/rai/ -
/screenshots/
-
If a notebook produces a dashboard, document:
-
model name/version
-
dataset used
-
compute used
-
dashboard configuration (e.g., class labels, sensitive features)
Where privacy techniques are demonstrated:
-
Use only non-sensitive / synthetic data in this repository.
-
Avoid committing any customer data, IDs, secrets, or endpoints.
Recommended:
-
Put privacy notebooks under
notebooks/dp100-azure-ml/privacy/ -
Put library notes and examples under
guides/orcheatsheets/
Typical components you'll see in the RAG notebooks:
-
ingestion (documents → chunks)
-
embeddings
-
AI Search index creation
-
retrieval + grounding
-
evaluation (quality + safety)
Store:
-
sample docs under
artifacts/sample-data/(only if redistributable) -
screenshots of indexes and evaluation under
artifacts/screenshots/
Fine-tuning notebooks (where supported) focus on:
-
dataset formatting
-
training job configuration
-
evaluation strategy (offline + human review checkpoints)
-
deployment notes
Important: Fine-tuning availability and exact steps can vary by tenant, region, and model.
Located under: cheatsheets/
This section is intended to be fast to scan and practical:
-
ML metrics: definitions + how to interpret + small examples
-
LLM evaluation: quality, groundedness, relevance, safety signals
-
Neural networks: MLP/CNN/RNN/GAN and "what they're good for"
-
Algorithms: simplified explanations for common models
Prefer saving models in MLflow format and storing only what's needed:
/models/<model-name>/...
/screenshots/
If you store larger model artifacts, enable Git LFS:
git lfs install git lfs track "*.pkl" "*.onnx" "*.pt" "*.joblib" "*.zip"
Tip: Avoid committing large, auto-generated outputs (logs, cache, temp files).
Contributions are welcome if they improve clarity and reusability:
-
Add explanations, diagrams, or troubleshooting notes
-
Improve notebook robustness (idempotency, better error handling)
-
Add "exam hints" sections (without copying protected exam content)
Suggested workflow:
-
Create a branch
-
Keep changes scoped (one notebook/guide per PR)
-
Add screenshots only when they add real value
-
Never include secrets or sensitive data
-
This repository contains original notes and extensions built on top of publicly available MicrosoftLearning labs.
-
Follow the licenses in the referenced MicrosoftLearning repositories for any reused content.
-
No affiliation with Microsoft is implied.
-
Do not upload proprietary, customer, or confidential data.
If you add a LICENSE file: MIT is common for your own content, but ensure it doesn't conflict with any upstream terms.
-
MicrosoftLearning Azure ML labs: https://github.com/MicrosoftLearning/mslearn-azure-ml
-
MicrosoftLearning AI Studio / AI Foundry labs: https://github.com/MicrosoftLearning/mslearn-ai-studio
If you find this useful, consider starring the upstream MicrosoftLearning repos as well.
-
Add
requirements.txt/environment.ymlaligned to the notebooks -
Add a
guides/setup-azure-ml.mdwith workspace + compute setup screenshots -
Add a
guides/setup-ai-foundry.mdwith project + connections setup -
Add
cheatsheets/ml-metrics.mdandcheatsheets/llm-metrics.md -
Add
.gitignorefor notebook outputs and local caches -
git clone https://github.com/boriszn/azure-ml-ai-foundry-labs-hub.git cd azure-ml-ai-foundry-labs-hub-
Create a virtual environment:
-
python -m venv .venv source .venv/bin/activate # macOS/Linux # .\.venv\Scripts\activate # Windows PowerShell -
Install deps (adjust as needed per notebook):
-
pip install -r requirements.txt -
Authenticate (if using Azure SDK/CLI workflows):
-