generative-ai-for-beginners

Delivers a 21-lesson progressive curriculum structured as 'Learn' (conceptual) and 'Build' (hands-on) modules that scaffold from LLM basics through advanced applications. Uses a modular Jupyter Notebook architecture with embedded code examples in both Python and TypeScript, allowing learners to execute concepts immediately within their development environment rather than reading static documentation.

Teaches prompt engineering through a two-tier approach: foundational techniques (clarity, specificity, role-based prompting) in Lesson 4, then advanced techniques (chain-of-thought, few-shot examples, system prompts) in Lesson 5. Each technique is demonstrated with concrete examples and code snippets showing how to structure prompts for OpenAI and Azure OpenAI APIs, with measurable improvements in output quality shown through side-by-side comparisons.

Lesson 10 teaches building AI applications using Azure AI Studio, a low-code/no-code platform that abstracts away API management and code complexity. Provides guided workflows for creating chat applications, search applications, and function-calling agents without writing code. Demonstrates how to configure models, define prompts, test interactions, and deploy applications through a visual interface. Enables non-technical users and rapid prototypers to build functional AI applications without software development expertise.

Lesson 2 teaches systematic model selection by comparing different LLMs (GPT-4, GPT-3.5, open-source models) across dimensions: cost, latency, quality, context window, and specialized capabilities. Provides a decision framework for choosing models based on use case requirements, with guidance on trade-offs between proprietary and open-source, larger and smaller models. Explains how to evaluate models empirically by testing on representative tasks rather than relying on marketing claims.

The curriculum is available in multiple languages (Chinese, Spanish, Portuguese, Japanese) with translations of all lessons and code examples. Each translation is maintained in the repository with language-specific directories, enabling learners to access the full course in their native language. Demonstrates commitment to global accessibility and removes language barriers for non-English speakers learning generative AI.

Provides a structured framework for responsible AI development covering bias detection, fairness assessment, transparency, and ethical considerations specific to generative AI. Lesson 3 integrates responsible AI practices as a foundational concept rather than an afterthought, with guidance on identifying potential harms, testing for bias in model outputs, and implementing safeguards. Uses Microsoft's responsible AI principles as the pedagogical framework.

Provides executable code examples and architectural patterns for building six distinct types of generative AI applications: text generation (Lesson 6), chat/conversational (Lesson 7), semantic search (Lesson 8), image generation (Lesson 9), low-code/no-code (Lesson 10), and function-calling-integrated (Lesson 11). Each lesson includes working code in Python and TypeScript that connects to actual APIs (OpenAI, Azure OpenAI, DALL-E), allowing learners to build and deploy functional applications rather than just understanding concepts.

Lesson 8 teaches semantic search by explaining vector embeddings, similarity matching, and retrieval-augmented generation (RAG) concepts, then provides code examples showing how to embed documents, store them in vector databases, and retrieve relevant context to augment LLM prompts. Lesson 13 (Advanced Topics) goes deeper into RAG patterns, vector database selection, and chunking strategies. The curriculum explains the architectural flow: documents → embeddings → vector store → retrieval → LLM context augmentation.

Lesson 14 (Advanced Topics) covers open-source LLM alternatives (Hugging Face models, Llama, Mistral) and their trade-offs versus proprietary APIs. Lesson 15 teaches fine-tuning approaches: parameter-efficient methods (LoRA, QLoRA) and full fine-tuning, with guidance on when each is appropriate. Provides code examples showing how to load open-source models, prepare training data, and execute fine-tuning workflows using libraries like Hugging Face Transformers and PEFT.

Lesson 12 teaches UX design principles specific to AI applications: handling uncertainty and hallucinations, designing for transparency, managing user expectations, and providing feedback mechanisms. Covers patterns like confidence scores, source attribution, fallback responses, and progressive disclosure of AI limitations. Provides design guidance rather than code, focusing on how to structure user interactions to account for LLM unreliability and the need for human oversight.

Lesson 16 (Advanced Topics) covers security considerations for AI applications: prompt injection attacks, data privacy in API calls, model poisoning, and securing API keys. Provides threat modeling guidance specific to generative AI systems, explaining attack vectors like adversarial prompts designed to bypass safety guidelines, and mitigation strategies like input validation, rate limiting, and secure credential management. Emphasizes that security in AI applications requires both traditional software security (API key management) and AI-specific concerns (prompt injection).

Lesson 11 teaches function calling (also called tool use) as a pattern for extending LLM capabilities by defining external functions the model can invoke. Provides code examples showing how to define function schemas, handle model-generated function calls, and execute them, using OpenAI's function calling API as the primary example. Explains the architectural pattern: user query → LLM generates function call → application executes function → result fed back to LLM → final response. Enables building agents that can interact with external systems, APIs, and databases.

Lesson 9 teaches image generation using DALL-E API, covering prompt engineering for image generation (different from text prompts), image editing and variation capabilities, and integration into applications. Provides code examples showing how to call the DALL-E API, handle generated images, and build workflows that combine text and image generation. Explains the differences between image generation, editing, and variation endpoints, and when to use each.