Generative AI Fundamentals

This four-module course gives you a clear, practical foundation in Generative AI from what it is and where it’s used, to how modern models work and how to apply them responsibly. You’ll start with the big picture: GenAI capabilities across text, image, audio, and video, plus real-world industry applications. Then you’ll dive into the science behind today’s Large Language Models: text representation (tokenization, embeddings), and the Transformer architecture (positional encoding, self-attention, encoder/decoder flow). Next, you’ll get hands-on with LLMs and workflows: crafting effective prompts, calling models via web/UI and APIs, running models locally (e.g., via Ollama), and extending capabilities with Retrieval-Augmented Generation (RAG) and fine-tuning. Finally, you’ll examine challenges and responsible practice, including copyright, privacy and security, explainability, and questions of ownership in the GenAI era. Designed for learners with basic Machine Learning and Python familiarity, the course blends short lessons with labs, quizzes, and exercises. By the end, you’ll understand the core concepts and architectures behind GenAI with a strong sense in ethical and responsible use and GenAI limitations. By the end of this course, learners will be able to: Explain how generative AI spans text, image, audio, and video and assess real industry workflows where it creates value. Trace the evolution of language modeling from probabilistic/NLP approaches to Transformers, and justify why attention overcomes prior limitations. Understand tokenization and word embeddings, and reason about how these representations affect model behavior. Decompose a Transformer block and follow tensors, through self-attention, MLPs, and normalization to explain how representations are formed and refined. Operate LLMs via web UIs, APIs, and locally with Ollama to write minimal inference code and improve outputs using prompt patterns and get familiar with concepts of RAG and Fine-Tuning as possible next steps. Identify, analyze, and explain LLMs shortcomings such as bias, hallucination, ownership, and prompt injection by formulating user-level guidelines, organizational processes, and governance policies.

Generative AI for Audio and Images: Models and Applications offers an in-depth exploration of how modern generative models such as Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Transformers, and Diffusion models are used to create, manipulate, and enhance audio, image, and video content. Learners examine the architectures, training processes, and use cases of these models across different modalities, gaining both conceptual understanding and practical insights through hands-on activities. The course also highlights the ethical and societal implications of generative AI, including bias, transparency, intellectual property, and the challenges of deepfake technologies. By covering foundational theory as well as state-of-the-art approaches and applications, this course prepares learners to apply and develop generative AI creatively and responsibly for the audio and image modalities. By the end of this course, learners will be able to: Outline core concepts, challenges, and the history of AI-generated audio. Analyze important foundational audio generation models, such as variational and vector quantized autoencoders (VAE and VQ-VAE) Examine how these models integrate with the latest GenAI technologies to form hybrid, state-of-the-art transformer and diffusion-based audio generation systems, Study the architecture and functionality of Generative Adversarial Networks (GANs), and their variations. Implement and train GAN models for creating and enhancing visual content, Explore cutting-edge techniques such as diffusion models and transformers for image and video creation. Discuss the ethical considerations regarding generative AI for audio and images.