Course Syllabus

Module 1: Introduction to Generative AI Concepts (4h)

This module introduces the fundamentals of generative AI. Learners will explore what generative AI is and how it is used today – for example, how language models can generate text, images or code from a simple prompt. We cover LLM basics (transformers, embeddings) and examine real-world examples (e.g. chatbots, image generation). Key topics include prompt engineering principles and an overview of the AI solution lifecycle.

• Objectives: Understand core generative AI capabilities; explain how transformer language models learn and generate new content.
• Outcomes: Describe generative AI and its applications; explain how prompts control model outputs; and outline responsible AI considerations in generative systems. For instance, learners will practice writing prompts and recognize issues like bias or harmful content that require mitigation.

Module 2: Azure Fundamentals for AI (3h)

This module reviews essential Azure cloud concepts for AI projects. Learners will get hands-on with the Azure Portal, creating a free Azure account and exploring resource groups, virtual machines, and storage accounts. We discuss Azure regions, subscriptions, and role-based access control (RBAC) relevant to data and AI. The class also introduces Azure’s security and compliance features. This ensures beginners understand the environment before building AI solutions on Azure.

• Objectives: Navigate the Azure Portal; manage subscriptions and resource groups; provision key Azure services.
• Outcomes: Configure an Azure subscription (free trial or Azure for Students) and set up a resource group and compute instance for AI work. Understand basic cloud concepts (as required by AI-900), including IaaS vs. PaaS, which underpin scalable AI deployments.

Module 3: Azure Cognitive Services Overview (4h)

This module surveys Azure’s pre-built AI services (formerly “Cognitive Services”) that support generative scenarios. We cover Vision APIs (computer vision, OCR, image analysis), Speech APIs (speech-to-text, text-to-speech), Language APIs (text analytics, translation, conversational language understanding) and Knowledge APIs (QnA Maker, Language Understanding). Participants will build simple demos: e.g., use Azure Translator to preprocess text, or use OCR to digitize images. These services can feed or augment generative models – for example, using language understanding to refine prompts. We also mention Azure Databricks as a data processing platform for large datasets.

• Objectives: Identify key Cognitive Services and their capabilities; use vision/speech APIs for data preprocessing.
• Outcomes: Demonstrate using Azure AI Language to summarize or translate text, Azure AI Vision to extract text from images, and understand how these outputs can be inputs to generative models. Recognize that Cognitive Services preprocess data (e.g. extracting features) before feeding it into LLMs.

Module 4: Azure OpenAI Service (6h)

In this hands-on module, learners work with Azure OpenAI Service to build generative solutions. Topics include signing up for Azure OpenAI access, exploring pre-trained models (GPT-4, GPT-3.5 Turbo, DALL·E, Embeddings), and using the REST or Python/SDK APIs. Students will create text generation and summarization pipelines, build a simple chatbot using chat completion, and generate images from prompts. We emphasize how Azure OpenAI brings enterprise security and compliance to OpenAI models. Lab exercises involve tuning model parameters (temperature, max tokens) and comparing model outputs on example prompts.

• Objectives: Deploy a GPT model in Azure; use completions and chat endpoints; generate images with DALL·E.
• Outcomes: Create a sample application (e.g. support chatbot) using GPT-4 on Azure. Explain Azure OpenAI’s model catalog and security features. Evaluate when to use different model sizes (e.g. GPT-4 vs GPT-3.5) for cost/performance tradeoffs.

Module 5: Azure Machine Learning for Generative AI (6h)

This module covers Azure Machine Learning (ML) for customizing and operationalizing generative AI. Topics include using Azure ML Studio, the Model Catalog (Foundation Models) and Prompt Flow. Participants will use Prompt Flow to automate and optimize prompt variants. We demonstrate fine-tuning a transformer model on custom data using Azure ML’s training clusters and GPU VMs, taking advantage of DeepSpeed and ONNX optimizations. We also cover MLOps: registering models, creating inference endpoints, and monitoring token usage and performance.

• Objectives: Use Azure ML workspace and compute resources; fine-tune a pre-trained model; deploy and monitor an inference endpoint.
• Outcomes: Train or fine-tune a generative model on Azure GPU VM and deploy it via Azure ML. Navigate the Azure ML model catalog to find open-source models. Build a CI/CD pipeline for model deployment and use metrics to evaluate model quality.

Module 6: Prompt Engineering & RAG Techniques (4h)

Effective prompting and data retrieval are key to production-quality generative AI. In this module we delve into prompt engineering strategies (system prompts, chain-of-thought, few-shot prompts) and demonstrate Retrieval-Augmented Generation (RAG). Learners will set up a vector database (using Azure Cognitive Search or Cosmos DB) and index a document collection. Then they implement RAG: the model retrieves context from data and uses it in prompts, improving accuracy. We illustrate this by building a question-answering “copilot” over a sample knowledge base.

• Objectives: Craft advanced prompts; implement RAG with Azure Search or Vector DB; tune prompt variability.
• Outcomes: Develop a RAG pipeline where GPT uses retrieved documents to answer queries. Run experiments to compare plain vs. RAG-enhanced responses. Cite Microsoft’s RAG module on Azure OpenAI as a guide.

Module 7: Building AI Agents and Copilots (3h)

This module explores AI agents (or copilots) – multi-turn bots that use generative AI to assist users. We discuss Azure AI Foundry/AI Studio for designing agents, and Azure Bot Service for chat interfaces. Learners will create a basic conversational agent that calls a GPT function (using “function calling”) or invokes an external API for up-to-date information. We cover state management in dialogues and integration with tools (e.g. calendar, knowledge API). Emphasis is on the latest Azure agent frameworks: according to Microsoft, “Azure OpenAI in Foundry Models” enables building cutting-edge AI agents with vision and language models.

• Objectives: Design a multi-turn chat agent; integrate GPT with an API or function call; deploy via Azure Bot or web app.
• Outcomes: Demonstrate a prototype AI assistant (e.g. booking bot or FAQ copilot). Explain how Azure AI services enable agent development. Understand the concept of “conversational” vs. “completion” modes in generative AI.

Module 8: Responsible AI and Content Safety (3h)

Generative AI must be used responsibly. This module covers Responsible AI principles (fairness, transparency, privacy) and Azure’s safety tools. We examine techniques to detect and mitigate harmful content in model outputs. Students use Azure AI Content Safety API to classify text/images and filter offenses (for example, NSFW or self-harm content). We also discuss data privacy, copyright issues, and governance: Azure’s compliance portfolio and data residency options. Guided labs include implementing content filters in a chatbot and applying red-teaming practices.

• Objectives: Learn ethical AI guidelines; use Azure Content Safety to scan outputs; apply bias detection.
• Outcomes: Apply severity scores from Azure’s Content Safety service to block or warn about unsafe output. Integrate fairness-check routines and document the solution’s compliance.

Module 9: Capstone Project & Labs (5h)

In this project phase, participants apply all skills to a real-world scenario. Teams (or individuals) choose a problem domain (e.g. customer support chatbot, document summarizer, image generation service) and build an end-to-end Azure solution. This involves data preparation, model selection/tuning, prompt design, and deploying the solution on Azure (web app or serverless function). Instructors provide sample datasets and use cases. Each project includes milestones (design, implementation, testing) and peer reviews. The goal is a working prototype and presentation.

• Objectives: Plan and implement a complete generative AI system on Azure; collaborate and iterate on design.
• Outcomes: Deliver a proof-of-concept solution (with code and documentation). Demonstrate use of Azure resources (OpenAI, ML, Cognitive Services) in the workflow. Receive instructor feedback on improving efficiency, security, and UX.

Module 10: Review & Certification Prep (2h)

The final module reviews key concepts and prepares learners for further certification or study. We revisit the main tools (Azure OpenAI, Azure ML, Cognitive Services) and discuss best practices. Knowledge check quizzes and a mock exam help consolidate understanding. We outline Microsoft certification paths: AI-900 (Fundamentals) and AI-102 (AI Engineer). As Microsoft notes, Azure AI-900 validates awareness of generative AI workloads, while AI-102 covers planning and implementing generative and agent solutions. Students learn how to access Microsoft Learn resources, exam guides, and possibly take free practice assessments.

• Objectives: Summarize the course topics; identify which Azure certification matches each skillset.
• Outcomes: Create a study plan for AI-900 or AI-102 if desired. Understand how the hands-on experience maps to exam domains (e.g. “Implement generative AI solutions”).

• Key Features

  • Hands-on Labs: Interactive exercises in the Azure Portal and notebooks. Learners build and deploy models in real-time. (Microsoft’s own AI Fundamentals course “includes hands-on exercises” to immerse learners in Azure AI services.)
  • Real-world Projects: A capstone project simulates a customer scenario (e.g. building a copilot for customer support). Students integrate multiple Azure services and datasets, applying their skills end-to-end.
  • Knowledge Checks: Each module ends with quizzes or assessments. For example, Microsoft Learn modules provide a “Module Assessment” that learners must pass. This course includes similar checkpoints to reinforce learning.
  • Certification Alignment (Optional): Module content is mapped to Microsoft Azure certifications. Topics align with Azure AI Fundamentals (AI-900) – which explicitly covers generative AI workloads – and Azure AI Engineer (AI-102), which tests implementing generative and agent solutions. Participants can opt to use this course as preparation for those exams.