Interactive AI learning lab

Learn AI by poking it.

Neuronauts turns hard AI ideas into small playable rooms. Drag sliders, run little experiments, and climb each idea from a plain-English giggle all the way up to the real equation.

Start at the Launchpad Browse the missions

New here? The Launchpad is your guided tour: it explains every mission and what you'll learn.

How every room works

One idea, four altitudes.

Each mission teaches the same concept four times, getting a little braver each step. Stop wherever it stops being fun, or push all the way to the math.

BeginnerPlain words, a funny everyday analogy, and a simple game.
IntermediateHow it actually works, with your first real worked example.
AdvancedThe real, properly typeset math, read back in plain English.
ExpertEdge cases, real-world usage, PyTorch, and a challenge quiz.

The map

Twelve rooms, organized into tracks.

Related rooms are grouped together. New here? Start at the Launchpad, then walk the Foundations, the two deep rooms everything else is built on. Want the full briefing and a step-by-step path? Read the Launchpad guide.

Start here

Orientation

Get the lay of the land first: what AI, ML, and deep learning even mean, and an animated map of how every room connects.

Start Launchpad The map of AI, and a guided tour of every other room. Learn: AI vs ML vs deep learning, neurons, softmax, the big picture. Open the guide →

Foundations · deep dives

The two pillars

Why it matters: nearly every modern AI system is a flavour of these two. They're the deepest, broadest rooms here. Machine Learning is the whole idea of learning from data; Deep Learning is the neural-network engine under GPT, diffusion, and robotics. Master these and the rest click into place.

Deep dive Machine Learning The big idea: learn patterns from examples instead of writing rules. Learn: supervised/unsupervised/RL, regression, classification, overfitting, k-means, SVM, PCA. Enter room → Deep dive Deep Learning Neural networks: many layers of neurons, the engine under modern AI. Learn: neurons, activations, backprop, CNNs, RNNs, GANs, autoencoders, train a net live - plus a frontier tier (with proofs): autodiff, gradient-descent convergence, momentum, Adam, initialization, double descent. Enter room →

Core building blocks

The mechanism everything reuses

Why it matters: every model turns data into vectors. (Gradient descent, how models learn, now lives in Deep Learning's frontier tier.)

Core concept Embeddings & Vector Search How words and images become arrows of meaning. Learn: vectors, embeddings, dot product, cosine similarity, vector search, plus a frontier tier on ANN indexes, contrastive learning & RAG. Enter room → Core concept Probability for Robotics A robot never knows its exact state, so it carries a belief and updates it with Bayes' rule. Learn: belief & the recursive Bayes filter, Gaussians, Kalman/EKF/UKF, particle filters, HMMs, information filter & SLAM - plus a research tier (with proofs & code): the free-energy principle, the Bayesian brain, energy-based models & learned differentiable filters. Enter room →

Language models & Attention

The transformer stack behind ChatGPT

Why it matters: this is the family powering today's LLMs and much of modern vision: self-attention, the transformer block, tokenization, generation, and the research that scales them up.

Transformers Transformers & LLMs Why every word looks at the other words, from self-attention to full language models. Learn: Q/K/V & attention math, multi-head, the transformer block, tokenization, sampling & the attention zoo, plus a frontier tier: RoPE, FlashAttention, Mixture of Experts, RLHF/DPO, with proofs. Enter room →

Generative & Embodied AI

Making things, and acting in the world

Why it matters: beyond predicting text, AI generates images and controls robots. These rooms build on the foundations: diffusion is a deep net that denoises; a robot is perception plus control in a loop.

Generative Diffusion Models How noise becomes a picture: denoising is generation. Learn: VAEs, ELBO, reparameterization, forward/reverse diffusion, guidance, plus a frontier curriculum (with proofs): normalizing flows, CNFs, flow matching, score/SDE, the EDM design space, latent diffusion, consistency models, and recent robotics (Diffusion Policy, π0, RDT, 3D Diffuser Actor). Enter room → Embodied Reinforcement Learning & Control AI with a body: sense, think, act, then learn. Learn: the sense-think-act loop, policies, control, feedback, plus a frontier tier (with proofs): Bellman/value iteration, policy gradients, Q-learning, PPO, LQR. Enter room →

Advanced Topics · research-grade

For when you want the frontier

Why it matters: deep, math-first dives into active research areas: with real derivations, step-through proofs you unfold yourself, animations of each method, and citations to the papers.

Advanced World Models for Robotics How robots learn to imagine, plan, and act, from Ha & Schmidhuber to V-JEPA 2. Learn: latent dynamics (RSSM/Dreamer), MPC (TD-MPC2), JEPA, diffusion world models, with proofs. Enter room → Advanced Vision-Language-Action Models How robots turn pixels + an instruction into motor commands, from behavior cloning to π0 and Gemini Robotics. Learn: imitation & the compounding-error theorem, action chunking (ACT), diffusion & flow-matching policies, RT-1/RT-2, OpenVLA, π0, FAST, with proofs. Enter room →

Useful Skills Wing

The craft around the AI

Why it matters: not AI-specific, but the practical foundation for any AI work: coding-interview pattern fluency and rock-solid Python.

Skill DSA Arena Thirty coding-interview patterns, with animated Python walkthroughs. Learn: the algorithm + variants for sliding window, graphs, DP, heaps, tries, union-find, and more. Enter room → Skill Python Playground Python fundamentals made visual, funny, and memorable. Learn: data types, loops, functions, OOP, generators, decorators, context managers, and more. Enter room →

Ready?

Climb your first idea from a giggle to an equation.

Start at the Launchpad