Learn how GenAI works, how modern AI models generate content, and how to build real-world GenAI applications — through interactive diagrams, visual workflows, and hands-on demonstrations.
A new class of artificial intelligence that doesn't just analyze or predict — it generatesGenerative — capable of producing new content (text, images, audio, code) rather than only classifying or predicting from existing data. entirely new content from learned patterns.
The broad field of building systems that can perform tasks normally requiring human intelligence — reasoning, perception, language understanding.
A branch of AI focused on creating models that produce new content — text, images, audio, video, or code — by learning distributions from training data.
Producing something new rather than classifying existing input. The model creates output that did not exist verbatim in its training data.
The fundamental pattern underlying every generative system
Trace how artificial intelligence evolved from rigid rule-based systems to today's multimodal generative models. Click any stage to explore it in depth.
Different models are optimized for different output modalities. Each card opens a detailed breakdown.
How generative models transform between different types of data
Different mathematical engines power different types of generation. Click an architecture to see how it generates data.
LLMs are the most impactful GenAI models today. Understand their types, deployment models, and the Transformer architecture that powers them.
A Large Language Model (LLM) is a neural network trained on massive text corpora to predict the next token in a sequence. Through this simple objective, models learn grammar, facts, reasoning abilities, and coding skills.
The core task: assign probabilities to sequences of text. Modern LLMs do this autoregressively — predicting one token at a time, conditioned on all previous tokens.
The engine of modern LLMs. Hover over any word to see what it "attends" to.
1. Embeddings: Convert tokens into high-dimensional vectors capturing meaning.
2. Positional Encoding: Add order information since attention is permutation-invariant.
3. Multi-Head Attention: Each token computes relationships with every other token. Multiple "heads" capture different relationship types (syntax, coreference, etc.).
4. Feed-Forward Network: Processes the attention output non-linearly.
5. Residual Connections: Information skips around layers to prevent gradient loss.
Select a task to see the suitable model type, expected input, output, and a typical workflow.
The complete workflow from problem identification to continuous improvement — including specialized patterns like RAG and fine-tuning.
Click any stage for explanation
GenAI works best as an assistant that amplifies human judgment — not as a replacement for it. Each side brings distinct strengths.
Traditional AI predicts; generative AI creates. Understanding the difference is foundational.
GenAI brings real productivity gains — and real risks. Click any card to reveal the mitigation strategy.
Real-world applications organized by domain. Select an industry to see concrete use cases.
A roadmap of emerging directions — distinguishing what works today from what's actively being developed.
Experiment with different prompt engineering techniques and see how the same task produces different outputs.
Click any stage for technical detail
Click any term to expand its definition