AI-900/AI-901 Azure AI Fundamentals Exam Preparation

Explore Microsoft’s six guiding principles for responsible AI, and examine unintended consequences, legal constraints, explainability, and societal harms from biased data and privacy concerns.

Learn the six Microsoft responsible AI principles—fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability—and how fairness differs from inclusiveness and transparency from accountability.

Explore Microsoft's responsible AI principles of inclusiveness, transparency, and accountability through examples like voice assistants, accessibility, and high-stakes decision making in hiring and loans.

Examine five fairness considerations for AI: training data bias, performance parity, sensitive feature leakage, allocation versus quality of service, and feedback loops, to ensure equitable outcomes across protected demographics.

Explore reliability and safety in AI systems by testing edge cases, understanding failure modes, defending against adversarial inputs, implementing human overrides, monitoring deployment drift, and applying guardrails for harmful content.

Explore privacy and security in AI by applying data minimization, consent, training-data privacy, defense against extraction attacks, secure infrastructure, synthetic data, and compliance with HIPAA, GDPR, PCI DSS.

design ai with inclusiveness at every step by ensuring accessibility standards, multilingual support, low-bandwidth options, representative training data, participatory design, and inclusive default settings.

Describe transparency in AI by explaining decisions (global and local), providing model cards with capabilities and limits, and signaling AI use. Log decisions, model versions, and uncertainties to support outcomes.

Explore accountability in AI solutions, including named owners, human review and escalation, governance, audit trails, and vendor accountability within Microsoft's responsible AI framework.

Explore how generative AI models work, from deep learning and neural networks to transformer architectures, foundation models, pretraining, and reinforcement learning, with prompts and probabilistic outputs.

Define the task and identify the appropriate AI model by capabilities, considering generative, classification, regression, clustering, vision, NLP, and speech tasks, plus pre-built versus custom options and safety.

Explore model deployment options across cloud-based services, managed services, and custom environments, exposing endpoints with authorization keys and tuning latency, scalability, content filters, and monitoring.

Explore Microsoft Foundry deployment options for GPT-4o, including model cards, token limits, older versions, and Playground testing with max tokens, temperature, and Top P.

Identify scenarios for common AI workloads by mapping tasks to models, from generative text creation to agentic automation, NLP, speech, computer vision, and information extraction, while considering cost and privacy.

Explore common text analysis techniques in natural language processing, including keyword extraction, entity detection, sentiment analysis, and summarization on unstructured text.

Identify features of speech recognition and speech synthesis, including speech-to-text conversion, real-time transcription, voice differentiation, and natural-sounding text-to-speech across languages and accents.

Identify features of computer vision and image-generation models in Azure. Show vision services analyze images and video, detect faces and text, and generative models create or modify images from prompts.

Extract information from text, images, audio, and videos using unstructured data techniques and AI document intelligence to convert invoices, forms, receipts, and other documents into structured data.

Learn to create effective system and user prompts for generative ai models, understand prompt engineering, context, constraints, and how to break tasks into steps using foundry's playground.

Create and test a single-agent solution in the Foundry portal by configuring an orchestration agent with a system prompt, optional tools, and an endpoint.

Create a lightweight python client for an ai agent using azure ai projects, configuring the agent name and version, and calling responses.create with user prompts in a loop.

Build a lightweight Python tool to perform text analysis, including summarization, named entity recognition, and key phrase extraction, using Azure AI Text Analytics and a deployed language service.

Respond to spoken prompts with a deployed multimodal model that accepts audio input and delivers audio or text outputs. Leverage Azure speech-to-text and a GPT-style service with endpoints and keys.

Build a lightweight app using Azure Speech in Foundry Tools to perform speech to text, configure the endpoint and key, install azure-cognitiveservices-speech, and compare SpeechRecognizer with SpeechSynthesis.

Learn to interpret visual input in prompts using a deployed multimodal model to describe images and extract visible text with OCR and Azure AI Vision.

Learn to create high-quality visuals by turning text into images with generative models, using descriptive prompts and the MAI image model to produce royalty-free, cinematic outputs.

Learn to extract information from audio and video with content understanding, producing timed transcripts, topic tags, sentiment, and insights on generative AI, neural networks, transformers, and prompt engineering.

Build a Python app that uses Azure Content Understanding to extract invoice fields from PDF and image and summarize audio from MP3.

Explore the AI-901 wrap-up with the GitHub repo of sample code and explanations to follow along. Master responsible AI concepts, large language models, and connecting with services using Python SDKs.

Explore common AI workloads on Azure, including prediction, anomaly detection, computer vision, natural language processing, knowledge mining, content moderation, and generative AI via Azure OpenAI, with supervised learning.

Explore the guiding principles for responsible AI, including fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability, and examine unintended consequences and bias in real-world systems.

Discover how fairness, reliability and safety, and privacy and security guide ai development, from fair triage and biased banking examples to edge-case testing and privacy concerns.

Explore how inclusiveness, transparency, and accountability drive AI development to ensure fairness, reliability, safety, privacy, and security, while supporting accessible decisions with human oversight in high-stakes contexts.

Explore Microsoft's AI principles and responsible AI resources, including six EHI principles and design guidelines with cards that show what the system can do, how well, and context-based timing.

Explore common machine learning types in Azure: regression for numerical predictions, classification with binary and multi-class options, and unsupervised clustering for insights, with examples like movie recommendations and spam detection.

Explore deep learning, a subset of machine learning, using layered neural networks on large data. See how it enables tasks like image recognition, speech processing, and language translation.

Explore the transformer architecture, including its encoder, decoder, and attention mechanism, and how it powers generative AI like GPT with contextual understanding and efficient translation.

Define features as input variables and labels as the prediction target, and apply the Goldilocks effect to choose just the right features for a life expectancy example.

Split your data into training and validation sets with random, representative samples, then select and evaluate regression, clustering, and classification algorithms in Azure Machine Learning using mean square error.

Evaluate classification models using a confusion matrix to compare predicted versus actual outcomes. Understand accuracy, precision, recall, and F1, plus training and testing data and business-context tradeoffs.