Industry-Relevant GenAI Course

Outcome: This week will help you understand the current business landscape of Generative and Agentic AI, while equipping you to identify key industry trends, opportunities, and challenges in their adoption.

Topics Covered:

• Overview of Generative AI: Definition, evolution, and impact on various industries.
• Overview of Agentic AI: Definition, characteristics, and potential applications.
• Market trends and business opportunities in GenAI and Agentic AI.
• Case studies of successful GenAI and Agentic AI implementations across different sectors

Outcome: This week will equip you with the fundamentals of prompt engineering across text, speech, and image modalities, teach you techniques to craft effective prompts for achieving precise AI model outputs, and help you understand the nuances of designing prompts for different Azure AI services.

Topics Covered: 

Outcome: This week, you will learn to develop Python scripts for interacting with Azure OpenAI and handling text data (Coding Track), as well as use Azure PromptFlow to design, test, and deploy prompt-based applications without extensive coding (No-Code Track)

Topics Covered: 

Coding Track:

• Text in Python: Strings, common string methods
• File Handling: Reading/writing files, working with .txt, .csv, .json
• Text Manipulation & Cleaning: Advanced string operations (find/replace, extract, split/join), removing unwanted characters, handling dates and numbers in text

No-Code Track:

• Promptflow Basics: Introduction, key features, setup, and integration
• Navigation & Workflows: Exploring the interface and building basic workflows

Sample Case Study: Cleaning and Analyzing Survey Responses at HealthTrack

Tool & Techniques Covered: 

Code: Colab, Azure AI Services/ Foundry, Azure OpenAI Model, Azure AI Inference client library (Python), POML/ prompty

No Code: Azure Prompt Flow

Outcome: This week enables you to leverage LLMs for business intelligence and automation by applying them to various natural language processing tasks, implementing text-to-text solutions such as summarization, translation, and content generation, and performing text-to-label tasks including sentiment analysis and entity recognition.

Topics Covered: 

• Explore business use cases for text classification and implement LLM-based classification workflows.
• Evaluate sentiment and aspect-based classification using accuracy metrics.
• Understand business applications of text generation and implement LLM-driven text-to-text workflows.
• Assess text generation outputs using ROUGE, BERTScore, and LLM-based evaluation.

• Generate automated summary reports and personalized product recommendations for customers

Tool & Techniques Covered: 

Code: Colab, Azure AI Foundry, Azure OpenAI models, Semantic Kernel SDK, POML

No Code: Azure Prompt Flow

TechGizmo, a leading tech company, needed a way to analyze thousands of customer reviews across multiple platforms to identify which product features, like performance, design, and usability, resonated with users, moving beyond general sentiment analysis.

Outcome: This week, you will learn to enhance AI solutions using embeddings, vector databases, and LLMs to build efficient Retrieval-Augmented Generation (RAG) systems for effective question answering.

• Explore business use-cases for RAG
• Build RAG workflows using LLMs
• Learn embeddings and how to choose embedding models
• Overview of vector databases
• Manage vector databases (CRUD operations)
• Implement RAG for question answering

Tool & Techniques Covered: 

Colab, Azure AI Foundry, Azure OpenAI Models, Azure AI Search

Outcome: This week, you will learn to implement Retrieval-Augmented Generation (RAG) for multimodal inputs, evaluating RAG model outputs, and mastering advanced techniques to improve the quality and performance of RAG systems in business applications.

Topics Covered:

• RAG with multimodal data sources (data with text, images and tables)
  
• Evaluating RAG outputs (Ragas and RAG Triad)
• Advanced Retrieval Mechanisms ( Hypothetical Questions, Query Expansion)
• Advanced Generation Mechanisms (Reranking, Information Compression)

Outcome: In this week, you'll learn to deploy and serve Large Language Models, optimizing them for real-world use. You'll also gain a thorough understanding of the end-to-end AI application workflow, covering prompt engineering, building interfaces with Streamlit, containerizing with Docker, and hosting models. 

Topics Covered:

• Master the deployment, management, and serving of Large Language Models (LLMs)
• Gain expertise in configuring and optimizing models for high-volume, high-performance tasks
• Use Streamlit to design user-friendly interfaces for AI applications
• Learn to Dockerize applications for seamless containerization and deployment

Outcome: This week covers the key differences between AI agents and LLM assistants, the evolution and use cases of generative AI agents, and the core functional modules of AI agent architecture. You will learn agentic design patterns, explore single-agent systems, and gain hands-on experience in designing and building simple autonomous AI agents.

• Generative AI Agents: Definition, use cases, and LLM vs. agent workflows.
• LLM Workflows with LangGraph/CrewAI: Sequential, parallel, and conditional workflows.
• AI Agent Architecture: Key components - planning, reasoning, execution, and memory.
• Types of Agents: Reflex, model-based reflex, goal-based agents.
• Agent Building: Python (LangChain with Azure) and no-code (Azure AI Agent Service)

Sample Case Study: Generative AI Agent for Automated Lead Generation and Customer Engagement

Tools & Techniques Covered:

Colab, Azure AI Foundry Agent Service (Agent Factory), Semantic Kernel Framework/ AutoGen Framework

This project involves deploying a Generative AI Agent to automate customer support for an e-commerce platform. The AI will handle common queries, process return requests, and provide personalized recommendations. It covers the deployment, optimization, containerization, and UI design of a high-performance support agent capable of managing high volumes of customer interactions.

Outcome: This week, you will learn foundations of LLMOps, focusing on the principles and practices of managing the lifecycle of large language models (LLMs). You will learn about model versioning, deployment strategies, and how to effectively manage LLMs

Topics Covered:

• Introduction to MLOps and LLMOps: Differences and similarities. 
• LLM lifecycle: Experimentation, training, deployment, monitoring. 
• Model versioning and registry in Azure Machine Learning. 
• Deployment of LLMs as endpoints (Coding Track) and through PromptFlow (No-Code Track). 

Outcome: This week, you will learn monitoring, logging, and performance tracking for LLMs, strategies for CI/CD in LLM applications, and key security and governance aspects in LLMOps.

• Monitor LLM performance: Latency, throughput, accuracy.
• Logging and observability for LLMs in Azure.
• CI/CD pipelines for LLMs (Coding Track) and automated deployments with PromptFlow (No-Code Track).
• Security best practices for LLM deployments on Azure.

Tools & Techniques:

Azure AI Foundry, Azure AI | Machine Learning Studio, Azure AI Pipelines

A digital marketing agency used LLMs to automate ad copy, blog posts, and social media content but struggled with managing multiple models, ensuring quality, and deploying updates. To improve efficiency and streamline deployments, they want to adopt LLMOps practices for monitoring, logging, and model management.

In this masterclass, you'll learn to overcome standalone model limitations by incorporating external context, understand the MCP architecture (Host-Client-Server) and its components, and explore benefits like enhanced accuracy and real-world use cases, all through practical, hands-on exercises.

Outcome: 

• Common terminologies used in NLP (Corpus/Vocabulary/Out-of-Vocabulary)
• Text Data and the Need for NLP (The Purpose of Text Data & NLP/Challenges with Text Data)
• Text Preprocessing: Common Methods (Treating accented and special characters/Lowercasing/Stop word Removal/Tokenization/Stemming & Lemmatization)
• Text Vectorization (Bag of Words (BoW)/Term-Frequency Inverse Document Frequency (TF-IDF))
• The ML NLP Pipeline (Text to Prediction)
• Distinguish between Base LLMs and Chat LLMs
• Understand the training mechanism of Large Language Models
• Gain familiarity with the multi-step GPT assistant pipeline used by the modern Chat LLMs of Generative AI

Outcome:

• Introduction to ML
• Supervised Learning VS Unsupervised Learning
• Classification vs Regression
• Role of Evaluation Metrics in ML
• Image related common terminologies used in CV (Pixel/Image channels/RGB Image)
• Image processing techniques (Edge Detection/Image Filters/The Kernel - Feature extraction)
• Introduction to CNN (Convolutional Layer/Multiple Convolutional layers/Padding & Pooling in CNN)

Outcome:

• Introduction to Responsible AI: Definition and importance in today’s AI ecosystem – Ethical AI vs. Responsible AI
• Core Principles of Responsible AI: Fairness & Inclusivity, Transparency & Explainability, Accountability & Governance, Privacy & Security
• Risks & Vulnerabilities in LLMs: Understanding LLM Limitations, Bias & Fairness Issues, Security Vulnerabilities
• Mitigation Strategies: Bias Mitigation, Hallucination & Reliability Control, Security & Safety Mitigation, Governance & Compliance

Outcome:

• Introduction to Prompt Refinement & DsPy
• Core Concepts of DsPy for Prompt Refinement