Fazmin / AILearningGuide

# Discover AI: A Learning Guide
A desktop app that teaches you how modern AI works by letting you play with it.
There are 51 small hands-on labs. In each one you move a slider, click a word, or train a tiny model, and you watch what changes. Next to every lab there is a short lesson that explains what you just saw. You can read it in short, simple sentences, in plain everyday words, or in the more technical version.
**Try it in your browser**. There's nothing to install.
Watch the tour as a video for a sharper version.
## About
I wanted to create a single place for anyone to learn AI. One that runs on your own computer, keeps track of what you've learned, lets you challenge yourself when you're ready, and makes hard ideas easier to understand through visuals and hands-on play.
## What's inside
The 51 labs are split into six groups. You can go in order or jump to whatever you're curious about.
| Group | What you'll learn |
| --- | --- |
| **Start with the groundwork** (1 to 6) | What AI is, the math you need, data and features, classic machine learning, the training loop, and how we know a model works |
| **Build the foundations** (7 to 14) | Tokens and embeddings, a single neuron, stacking neurons, loss and gradient descent, backpropagation, training dynamics, why GPUs matter, and the main model types |
| **Look inside models** (15 to 21) | Next-token prediction, attention, the transformer block, the KV cache, attention at scale, scaling laws, and structured decoding |
| **Build with a trained model** (22 to 27) | Search with embeddings, RAG, tools and function calling, agents, multimodal models, and failure modes and safety |
| **Explore the frontier** (28 to 35) | Reinforcement learning, post-training, in-context learning and reasoning, diffusion, the AI ecosystem, ethics, and interpretability |
| **Train and adapt a model** (36 to 51) | A full training run from start to finish: datasets, synthetic data, hyperparameters, training from scratch, fine-tuning, LoRA, model merging, instruction tuning, preference tuning, reasoning models, evaluation, quantization, distillation, serving, and a final "ship it" capstone |
## Screenshots
**Learning map.** The home screen shows all six groups of labs and how far you've got.
**Attention lab.** Click a word and see which other words it looks at. This one runs a real small transformer right in the app.
**Backpropagation lab, in dark theme.** Step through the forward pass and then follow the gradients backward, one calculation at a time.
**Train one from scratch.** Pick some text, a model size, and a training recipe. The model retrains on every change, so you can watch the loss fall and the samples get better.
**Connections map.** See how all 51 labs link together. Pick one to see what it builds on and what it leads to.
**Simple, Plain or Standard.** The welcome tour shows the same idea written all three ways, so you can pick the one that suits you.
## Features
- **51 interactive labs.** Each lab is split int