EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content

Interactive Explainers

See the concept, don't just read about it.

These are the best interactive visualizations available for core AI/ML concepts - curated, categorized, and explained so you know what you're looking at before you open them.

We are also building our own explainers for this platform. Watch this space.

Transformers & Attention​

Transformer Explainer​

poloclub.github.io/transformer-explainer Watch a real GPT-2 model process text token by token. See embeddings, attention heads, residual connections, and the softmax output - all animated, all interactive. The single best tool for understanding how a transformer works. Best for: Anyone learning transformers for the first time. Also excellent for explaining transformers to non-ML engineers.

BertViz​

github.com/jessevig/bertviz Visualizes attention patterns in BERT, GPT-2, and other transformers. Head view (individual attention heads), model view (all heads across all layers), and neuron view (individual neuron activations). Best for: Understanding what attention heads actually learn. Revealing when you fine-tune a model and want to see what changed.

Embeddings & Semantic Space​

Embedding Projector (TensorFlow)​

projector.tensorflow.org Visualize high-dimensional embeddings in 3D using PCA, TSNE, or UMAP. Load your own embeddings or use the preloaded Word2Vec/GloVe datasets. Best for: Understanding semantic clustering. Verifying that your embeddings separate concepts correctly. Debugging embedding quality.

Nomic Atlas​

atlas.nomic.ai Upload a dataset with text, get an interactive 2D map of semantic space. Navigate millions of points. See clusters, outliers, and concept neighborhoods. Best for: Exploring large document corpora. Understanding what your RAG index actually contains. Finding gaps in training data.

Diffusion Models​

Diffusion Explainer​

poloclub.github.io/diffusion-explainer Step through the diffusion process - watch noise become an image, one denoising step at a time. See the role of the U-Net and time embeddings. Best for: Understanding stable diffusion from first principles. From the same team as the Transformer Explainer.

Neural Networks (Fundamentals)​

Neural Network Playground (TensorFlow)​

playground.tensorflow.org Train a neural network in your browser on 2D datasets. Adjust layers, neurons, activation functions, regularization - and watch decision boundaries form in real time. Best for: Intuition for how depth, width, and activation functions affect learning.

Backpropagation Visualizer - 3Blue1Brown​

3blue1brown.com/topics/neural-networks Not interactive but the clearest explanation of backprop, gradient descent, and the chain rule that exists. Essential viewing before reading any optimization paper.

Reinforcement Learning​

RL Gym Visualizer​

gymnasium.farama.org Standard RL environments with visual output. Run standard algorithms and watch agents learn in real time. Best for: Intuition for how reward signals shape behaviour. Required for reading RLHF papers with real understanding.

Coming from EngineersOfAI​

We are building our own explainers for:

  • RAG Pipeline - watch a query flow through retrieval, reranking, and generation
  • Attention Deep Dive - multi-head attention, KV cache, rotary position embeddings
  • Agent Loop - ReAct reasoning trace, tool calls, and context management
  • Embedding Space - semantic similarity, cosine distance, approximate nearest neighbours

Subscribe to AI Letters to be notified when these go live.

© 2026 EngineersOfAI. All rights reserved.