Tools
Small interactive demos that illustrate ideas from my teaching and research. Everything runs in your browser — nothing is sent anywhere.
Word2Vec Simulator
See how words become vectors — and why 'king − man + woman ≈ queen'.
An interactive playground that builds a small word embedding space in your browser. Explore neighborhoods, run vector arithmetic, and watch meaning emerge from geometry.
Open the demo →Transformer Residual Stream
A minimal 'Attention Is All You Need' encoder you can poke at — watch the stream change layer by layer.
An interactive encoder block from the original Transformer paper. Each layer writes twice to the residual stream — once from self-attention, once from the feed-forward network. Toggle residual connections or layer norm to see what breaks.
Open the demo →LIME Explainer
Why did the model say that? LIME builds a local, human-readable explanation.
An interactive LIME simulator. Pick a point on a 2D decision surface, watch the perturbation cloud form, and see the weighted local linear model that explains the black box's prediction at that point.
Open the demo →SHAP Explainer
Fair feature attribution via Shapley values — local explanations and global importance in one tool.
An interactive SHAP simulator with two views. Local: pick an instance, see the exact Shapley values, the force plot, the waterfall, and every coalition the formula averages over. Global: mean |SHAP| importance and a beeswarm plot across hundreds of instances.
Open the demo →Cosine Similarity Explorer
Drag two vectors and watch similarity change — direction matters, magnitude doesn't.
An interactive 2D playground for building intuition about cosine similarity. Drag vector endpoints to see how angle determines similarity while length is ignored — the key insight behind how word embeddings measure meaning.
Open the demo →