Labs

25 hands-on AI labs that run real algorithms in your browser: neural networks, gradient descent, A* pathfinding, attention, diffusion, clustering and more.

Labs

• Neural Network Playground

  Build a network layer by layer and watch it learn a decision boundary in real time.

• Gradient Descent Lab

  Drop optimizers onto a loss landscape and race SGD, momentum, RMSProp and Adam to the minimum.

• Decision Tree Builder

  Grow a decision tree split by split and see how impurity guides every choice.

• Clustering Visualizer

  Run K-means and DBSCAN step by step on blobs, moons and rings.

• Confusion Matrix Explorer

  Slide a decision threshold and watch precision, recall and the ROC curve trade off.

• Fairness Lab

  Set loan-approval thresholds for two groups and discover why every fairness metric can’t be satisfied at once.

• Image Filter Playground

  Apply convolution kernels pixel by pixel and see the edge detectors CNNs learn on their own.

• Object Detection Lab

  Run a real COCO-SSD detector in your browser on your camera or sample images.

• Text Playground

  Tokenize text, score sentiment and explore how words become vectors.

• Recurrent Networks Lab

  Step a recurrent network through a sequence and watch its memory fade or hold.

• Attention Visualizer

  Compute queries, keys and values by hand and see which words attend to which.

• Transformer Explorer

  Follow a sentence through embeddings, attention, feed-forward layers and out as a prediction.

• Diffusion Lab

  Add noise to an image step by step, then run the process in reverse to generate one.

• Pathfinding Visualizer

  Draw a maze and watch A*, Dijkstra, BFS and DFS explore it cell by cell.

• Travelling Salesman

  Place cities and pit greedy search, 2-opt, simulated annealing and genetic algorithms against each other.

• Reinforcement Learning Arena

  Train a Q-learning agent in a gridworld and watch its value map form episode by episode.

• Next-Token Sampling Lab

  Train a tiny language model in your browser, then steer its output with temperature, top-k and top-p.

• Embedding Explorer

  Embed a set of documents, project them to 2D, and run a semantic search you can inspect step by step.

• Calibration Lab

  Does 80% confident mean right 80% of the time? Build a reliability diagram and fix an overconfident model.

• Boosting Lab

  Add trees one at a time and watch each new one fix the mistakes of the ensemble so far.

• Recommender Lab

  Rate a few films, train matrix factorization live, and see your taste as a point in the same space as the films.

• Model Compression Lab

  Shrink a trained network with low-rank factorization and quantization and see exactly what accuracy it costs.

• Tokenizer Lab

  Train a byte-pair tokenizer on your own text, watch merges form, and see why models stumble on spelling and arithmetic.

• Reward Model Lab

  Judge pairs of answers, train a reward model on your preferences, then watch a policy optimise it and start to game it.

• Mixture of Experts Lab

  Train a small mixture-of-experts network live and watch the router carve the data between specialists.