Azure AI Fundamentals AI-900 Foundations

The Microsoft Azure AI Fundamentals AI-900 exam is an entry-level certification for people who want to describe AI concepts and Azure AI services. It is useful for business stakeholders, data professionals, and developers who need a common language around AI workloads. The exam covers how to describe AI workloads and considerations, fundamental principles of machine learning, features of computer vision workloads, features of natural language processing workloads, and features of generative AI workloads. AI workloads use models to interpret data: computer vision understands images, natural language processing understands text and speech, and generative AI creates new content. Responsible AI principles such as fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability guide how systems should be designed and used. This course pairs each topic with a practice bank so you can apply what you learn.

Machine learning lets software learn patterns from data instead of following fixed rules. In supervised learning, each training example has a known label, and the model learns to map features to that label. Regression predicts numeric values such as house prices, and classification predicts categories such as spam or not spam. Unsupervised learning finds structure without labels; clustering groups similar items together. Features are the input variables, and the label is the target output. During training, the model adjusts its parameters to reduce error. Validation and test data check whether the model generalizes to new examples. Overfitting occurs when a model memorizes training data and performs poorly on unseen data. Azure Machine Learning provides workspaces, experiments, AutoML, and endpoints for the full lifecycle. This lesson connects each concept to practice questions.

Computer vision workloads let systems interpret images and video. Image classification assigns one label to an entire image, such as cat or dog. Object detection finds individual objects and returns bounding boxes around them. Semantic segmentation labels every pixel so the system understands the shape and location of objects in detail. OCR reads printed and handwritten text from images and documents, which powers invoice processing, license plate reading, and searchable scans. Face detection finds faces and attributes such as pose, while face recognition identifies or verifies a person. Azure AI Vision provides prebuilt image analysis, OCR, and custom vision capabilities. Azure AI Document Intelligence extracts fields and tables from forms, contracts, and receipts. Practice by matching each scenario to the correct vision task.

Natural language processing lets systems read, understand, and generate human language. Sentiment analysis determines whether text is positive, negative, or neutral, which is useful for product reviews and customer feedback. Named entity recognition extracts people, places, organizations, and dates. Key phrase extraction returns the most important phrases in a document. Language detection identifies the language of input text. Azure AI Translator converts text between languages, while Azure AI Speech handles speech to text, text to speech, speech translation, and speaker recognition. Language understanding models capture user goals as intents and supporting details as entities. Custom question answering finds answers from documents and knowledge bases. Document intelligence complements NLP by extracting structured fields from forms. These capabilities combine to power chatbots, call center analytics, search, and translation applications.

Generative AI creates new content from learned patterns. Large language models are trained on large text corpora and can generate completions, summarize, translate, and answer questions. A prompt is the input instruction, and a completion is the generated output. Models process text in tokens, which affect cost and limits. Temperature controls randomness: higher values create more varied output and lower values create more focused output. Embeddings represent text as numeric vectors so similar meanings can be compared. Azure OpenAI provides models such as GPT and DALL-E. Retrieval augmented generation combines retrieved documents with a generative model so answers can be grounded in evidence. Responsible AI is essential for generative systems: content filters block harmful material, human review catches unsafe outputs, clear labeling discloses AI-generated content, and testing identifies bias. Fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability guide every deployment.