Microsoft Azure Artificial Intelligence Fundamentals (AI-900)

Exam-focused summary of the Microsoft Azure AI Fundamentals (AI-900) services and topics.

AI workloads and responsible AI

The exam tests recognizing common AI workloads — computer vision, natural language processing, document processing, and generative AI — and when to use each in real scenarios. It emphasizes Microsoft’s responsible AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.

Artificial Intelligence workloads > Vision, NLP, Document processing, Generative AI > Choose vision for images, NLP for text/speech, document intelligence for forms/invoices, generative AI for summarization/chat/content.

Responsible AI principles > Fairness, Reliability & Safety, Privacy & Security, Inclusiveness, Transparency, Accountability > Apply content filters, guardrails, data protection, explainability, human oversight in solutions.

Machine learning fundamentals on Azure

Know supervised vs. unsupervised learning and typical use cases: regression (predict numbers), classification (predict categories), clustering (grouping), plus basics of deep learning and the Transformer architecture. Understand core ML concepts: features vs. labels, training/validation/test splits, model evaluation at a high level, and where Azure Machine Learning helps with data/compute, AutoML, model management, and deployment.

ML techniques > Supervised (regression/classification), Unsupervised (clustering), Deep learning, Transformers > Predict prices (regression), spam/not‑spam (classification), customer segments (clustering), language tasks with transformers.

ML concepts > Features/labels, train/validate/test splits, evaluation at a high level > Select relevant features, hold out validation, check accuracy/precision/recall at a concept level.

Azure Machine Learning > Data/compute, Automated ML, model registry, endpoints > Use AutoML for quick baselines, register and deploy models to real‑time endpoints.

Computer vision on Azure

Identify solution types: image classification (what’s in the image), object detection (what and where), optical character recognition/OCR (read text), and facial detection/analysis (find faces and assess attributes). Map tasks to services: Azure AI Vision (analysis, OCR), and Azure AI Face (face detection/analysis); know what they do at a high level.

Solution types > Image classification, Object detection, OCR, Face detection/analysis > Classify products, detect helmets in safety cams, extract text from receipts, detect faces for photo organization.

Azure tools > Azure AI Vision, Custom Vision (train own classifiers/detectors), Azure AI Face > Use OCR for bills/IDs, Custom Vision for bespoke label sets, Face for detecting faces and attributes.

Natural language and speech

Recognize NLP workloads: key phrase extraction, entity recognition, sentiment analysis, language modeling, speech recognition/synthesis, and translation. Map tasks to services: Azure AI Language (text analytics, Q&A, entities, sentiment) and Azure AI Speech (speech-to-text, text-to-speech, translation).

NLP tasks > Key phrase extraction, Entity recognition, Sentiment analysis, Language modeling > Summarize topics in reviews, pull organizations/people, gauge sentiment in social posts.

Speech tasks > Speech‑to‑Text, Text‑to‑Speech, Translation > Transcribe call center audio, generate IVR voices, translate multilingual chats.

Azure tools > Azure AI Language, Azure AI Speech > Use Text Analytics for key phrases/entities/sentiment; Speech service for STT/TTS/translation.

Generative AI on Azure

Understand what generative AI is, typical scenarios (content generation, summarization, chat), and responsible AI considerations. Know the platform pieces: Azure AI Foundry (hub for building, orchestration, safety) and Azure OpenAI Service (access to foundation models and capabilities) at a high level.

GenAI concepts > Foundation models, prompts, safety/guardrails > Build chatbots, summarizers, content generators with responsible usage.

Azure services > Azure OpenAI Service, Azure AI Foundry (hub, orchestration, model catalog) > Host GPT‑style models, manage prompt flows, content filters, and evaluations.

Scenarios > Summarize documents/emails, code assistance, knowledge chat over internal data > Add retrieval over enterprise content with grounding and citations.

Document intelligence

• Workloads > OCR, key‑value extraction, table extraction, layout understanding > Process invoices, receipts, IDs, forms at scale.
• Azure tool > Document Intelligence (formerly Form Recognizer) > Use prebuilt models for invoices/IDs; train custom models for unique forms.

Choosing the right service quickly

• Vision classification/detection > Custom Vision under Azure AI Vision > Custom product labeling, defect detection on a line.
• OCR and documents > Azure AI Vision OCR or Document Intelligence > Receipts, invoices, passports with structured outputs.
• Text analytics > Azure AI Language > Monitor sentiment and entities in feedback streams.
• Speech > Azure AI Speech > Meeting transcription and voice responses.
• ML training/deploy > Azure Machine Learning/AutoML > Train a churn model and deploy as a REST endpoint.
• Generative chat/summarize > Azure OpenAI + Azure AI Foundry > Internal knowledge chatbot with safety filters and logging.

Governance and security (exam awareness)

• Data protection > Responsible AI, content filters, PII handling > Redact sensitive data and apply safety filters before output.
• Monitoring and lifecycle > Model registry, versioning, evaluation > Track models, compare runs, roll back if performance drifts.

Edge cases and pitfalls

• Face usage limits > Policy and responsible use constraints > Ensure compliance and avoid prohibited uses.
• OCR vs Document Intelligence > Use OCR for simple text, Document Intelligence for structured extraction > Pick based on complexity and document variety.

7 day study plan and cheatsheet for Microsoft Azure AI Fundamentals (AI-900)