Cloud Pro: Azure AI and Machine Learning: Choose, Compare and Launch

What's inside

1. Level 1: Every Azure AI and ML Service on One PageFree
   Learners build a single grouped map of Azure's AI and machine learning services. The groups are generative AI platforms and models, agents, search and retrieval, ready-made AI APIs, machine learning platforms and accelerators, and each service is shown with its current and former names, such as Azure AI Studio to Azure AI Foundry to Microsoft Foundry, and Cognitive Services to Azure AI services to Foundry Tools.
2. Level 2: The Standard AI Workload ProfileFree
   Learners write a standard profile for any AI workload. It records the task type (generative, predictive or perception), whether to buy an API, adapt a model or train one, data sensitivity and residency, latency and throughput, volume and token or request costs, model choice and lock-in, evaluation and safety, MLOps maturity, accelerator needs, team skills and the cost model.
3. Level 3: Shared Groundwork: Regions, Quotas, Identity, Network
   Learners set up the foundations every later service relies on: checking which models and services each region offers, reading and requesting quota, using Microsoft Entra ID and managed identities instead of keys, and placing services behind private endpoints.
4. Level 4: Microsoft Foundry Resources and Projects
   Learners work through Microsoft Foundry as a full nine-part service lesson. It covers the Foundry resource and Foundry project model versus classic hub-based projects, what each one provisions and secures, its limits and costs, and a quick start that deploys a resource and project with the CLI and Bicep or Terraform, then cleans them up.
5. Level 5: The Foundry Model Catalogue and Azure OpenAI Models
   Learners explore the Foundry model catalogue: Azure OpenAI models, other models sold directly by Azure, and partner and community models. The lesson covers how each kind is licensed, hosted, billed and retired, with a quick start that deploys and calls one model.
6. Level 6: Deployment Types, Quotas and Provisioned Throughput
   Learners compare Foundry model deployment types: global, data zone and regional standard, batch, and provisioned throughput. They weigh data residency, tokens-per-minute quota, latency and cost, and work out when reserved capacity beats pay-as-you-go.
7. Level 7: Fine-Tuning Models in Foundry
   Learners decide when to fine-tune a model in Foundry instead of improving the prompt or adding retrieval. They prepare training data, run a fine-tuning job, and weigh the ongoing cost of training and of hosting the result.
8. Level 8: Foundry Agent Service
   Learners work through Foundry Agent Service as a full nine-part service lesson. It covers agents, threads and tools, connecting to knowledge and actions, its limits and costs, who operates what, and a quick start that creates and runs a minimal agent, then cleans it up.
9. Level 9: Azure AI Search: Indexes, Keyword, Vector and Hybrid
   Learners learn Azure AI Search from the ground up: indexes, indexers and skillsets, keyword versus vector versus hybrid queries, and the semantic ranker. A quick start builds a small hybrid index and queries it.
10. Level 10: Azure AI Search at Scale: Tiers and Cost
    Learners size Azure AI Search for production. They choose a service tier and set replicas and partitions for availability and throughput, apply storage and vector index limits, check AI enrichment and semantic ranker charges, and compare the result with other ways to do retrieval.
11. Level 11: Azure Speech in Foundry Tools
    Learners work through Azure Speech as a full nine-part service lesson. It covers speech-to-text (real-time, fast and batch transcription), text-to-speech, custom speech and voice, limits, pricing per audio hour and per character, and a quick start that transcribes a file, then cleans up.
12. Level 12: Azure Vision and Face in Foundry Tools
    Learners work through the Vision services as a nine-part lesson: image analysis, OCR (reading text in images), custom image models and the restricted Face service. It covers which features are retired or need an approved application, pricing per transaction, and a quick start that analyses one image.
13. Level 13: Azure Language and Translator in Foundry Tools
    Learners work through Language and Translator as a nine-part lesson: detecting personal data, summarisation, conversational language understanding, custom text classification, and document and text translation. They also judge when a general-purpose large language model now does the job better or cheaper.
14. Level 14: Document Intelligence in Foundry Tools
    Learners work through Document Intelligence as a nine-part lesson: read, layout and prebuilt models, custom extraction and classification models, page limits and file limits, pricing per page, and a quick start that extracts fields from an invoice.
15. Level 15: Azure AI Content Safety in Foundry Tools
    Learners work through Azure AI Content Safety as a nine-part lesson: text and image moderation, Prompt Shields against jailbreaks and prompt injection, groundedness detection and protected material checks. They also see how it relates to the content filters built into Foundry model deployments.
16. Level 16: Foundry Tools: Commitment Tiers, Containers, Retirements
    Learners cover the operational side of Foundry Tools: commitment tier pricing, running supported models in disconnected or connected containers, multi-service versus single-service resources, and the retirements. Metrics Advisor retired in May 2026, and Anomaly Detector and Personalizer reached their 1 October 2026 retirement date.
17. Level 17: Azure Machine Learning: Workspaces and Training
    Learners work through Azure Machine Learning for training: workspaces, compute clusters and serverless compute, environments, data assets, training jobs and pipelines, AutoML and the registry. They also see how it differs from Foundry for predictive work.
18. Level 18: Azure Machine Learning: Endpoints and Hosting
    Learners deploy trained models with Azure Machine Learning online and batch endpoints. They compare managed compute with Kubernetes-attached compute, route traffic between deployments with blue-green and safe rollout, and find the always-on hosting cost of each choice.
19. Level 19: Fabric Data Science in Brief
    Learners get a short, practical look at data science in Microsoft Fabric: notebooks, MLflow experiments and models, and batch scoring over OneLake data. The focus is on when a Fabric-centred data team should use it instead of Azure Machine Learning.
20. Level 20: GPU and Accelerator Options in Brief
    Learners survey Azure's GPU and accelerator options: the main GPU VM families, using them through Azure Machine Learning, Azure Kubernetes Service (AKS) and serverless GPUs in Container Apps, and quota and regional availability. The focus is on when renting your own accelerators beats a managed model deployment.
21. Level 21: Choosing and Evaluating Models
    Learn to narrow the Foundry model catalogue to a shortlist using the workload profile, then prove the choice with an evaluation set, automated and human-graded metrics, and repeatable comparisons instead of public benchmark scores.
22. Level 22: Designing Retrieval-Augmented Generation
    Design a production retrieval-augmented generation (RAG) pipeline end to end: chunking, embeddings, hybrid retrieval with semantic ranking in Azure AI Search, prompt grounding and citations. Then decide when RAG is the wrong tool compared with fine-tuning, long context or an agent.
23. Level 23: Responsible AI, Content Safety and Data Protection
    Apply a layered safety and data protection design across models, agents and AI APIs: content filters and prompt-injection defences, red-teaming, data handling and residency, identity, private networking and customer-managed keys. Map each layer to the responsible AI obligations in the profile.
24. Level 24: MLOps and LLMOps at the Right Maturity
    Match the operating model to the team's MLOps maturity, from manual notebooks to fully automated pipelines. Learn how versioning, CI/CD, registries, monitoring and drift or quality alerts work for both trained models in Azure Machine Learning and prompt-plus-model apps in Foundry.
25. Level 25: Worked Cost Model: A Generative Workload
    Build a dated, line-by-line cost model for a RAG chat assistant, covering input and output tokens, embeddings, Azure AI Search tiers and replicas, agent tool calls, logging and networking. Then compare pay-as-you-go with provisioned throughput and test how sensitive the total is to volume.
26. Level 26: Worked Cost Model: A Predictive Workload
    Build a dated cost model for a predictive workload on Azure Machine Learning, covering training compute, hyperparameter sweeps, storage, real-time versus batch endpoints and retraining cadence. Then compare it with buying a ready-made AI API for the same job.
27. Level 27: Building a Weighted Selection Scorecard
    Turn the standard workload profile into a weighted scorecard with explicit criteria, weights, scoring scales and knock-out rules, and use it to rank candidate Azure services for a real workload.
28. Level 28: Scoring Azure Against Another Cloud
    Apply the same scorecard to an equivalent service stack on another cloud (such as AWS or Google Cloud), collecting dated evidence from official docs and price pages, and weigh portability, lock-in and migration cost honestly.
29. Level 29: Final Challenge: Decision Workshop and Selection Record Exam
    Run a full decision workshop on a new scenario, covering the profile, shortlist, evaluation plan, safety and data design, MLOps approach, dated cost model, weighted scorecard and cross-cloud comparison. Finish with a one-page selection record you could defend to an architecture review board.

Access

The first 2 levels are free with a free account. Every level, the podcast edition and the AI tutor come with All Access at £4.99/month or any Creator plan — see pricing.

Start free →