Azure AI-900 Cheat Sheet
The Azure AI services, machine-learning concepts, and responsible-AI principles worth memorising for AI-900 — on one page.
Much of AI-900 is matching a scenario to the right Azure AI service and the right concept. Nail these and most questions become recognition, not recall.
Fairness (treat all groups equitably), reliability and safety (perform consistently and safely), privacy and security (protect data), inclusiveness (work for people of all abilities), transparency (be understandable), and accountability (people are answerable for the system). Memorise all six — Microsoft tests them by name and by scenario.
Machine-learning concepts
A model learns from features (the input columns) to predict a label (the answer). You split data into training and validation sets to measure how well it generalises.
| ML task | Predicts | Example |
|---|---|---|
| Regression | A numeric value | Predict a house price or tomorrow's temperature |
| Classification | A category or class | Is this email spam or not spam? |
| Clustering | Groups in unlabeled data | Segment customers into similar groups |
Azure AI services by workload
You need to know which service handles which workload, not how to build it.
| Workload | Azure service | Use it for |
|---|---|---|
| Build/train models | Azure Machine Learning | Automated ML and the designer for building custom models |
| Vision | Azure AI Vision | Image analysis, tagging, and optical character recognition (OCR) |
| Vision (faces) | Azure AI Face | Detect and analyse human faces in images |
| Language | Azure AI Language | Sentiment analysis, key phrase extraction, entity recognition |
| Translation | Azure AI Translator | Translate text between languages |
| Speech | Azure AI Speech | Speech-to-text, text-to-speech, and speech translation |
| Documents | Azure AI Document Intelligence | Extract fields and tables from forms and documents |
| Generative AI | Azure OpenAI Service | Access large language models for text and code generation |
| Search | Azure AI Search | Add AI-powered search and retrieval over your own content |
| Safety | Azure AI Content Safety | Detect harmful or unsafe content in text and images |
A large language model (LLM) predicts the next token (a chunk of text) given a prompt, producing a completion. Giving the model relevant context so it answers from your data is called grounding. LLMs are built on the transformer architecture. In Azure, generative AI is delivered through the Azure OpenAI Service and Azure AI Foundry.
- Six responsible-AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability.
- Regression predicts numbers; classification predicts categories; clustering groups unlabeled data.
- Match the workload to the service: Vision (images/OCR), Language (text), Speech, Document Intelligence (forms), Azure OpenAI (generative AI).
- Generative AI: LLMs predict tokens from a prompt to produce a completion; grounding adds your own context.
Frequently asked questions
What Azure service is used for generative AI on AI-900?
The Azure OpenAI Service (delivered through Azure AI Foundry) provides access to large language models for text, code, and image generation.
What is the difference between regression and classification?
Regression predicts a numeric value (such as a price), while classification predicts a category or class (such as spam or not spam).
Independent study resource. Not affiliated with, authorized, endorsed by, or sponsored by CompTIA, Amazon Web Services, Microsoft, or ISC2. All trademarks are the property of their respective owners and are used here for identification only. All practice questions are original.