Nerdy Stuff

Under the hood explorations. See how AI actually works with interactive visualizations.

Embedding Explorer

Live

Visualize how texts cluster in vector space using UMAP. See semantic similarity in action.

EmbeddingsUMAPTransformers.js

Rusenizer

Live

My Turkish-optimized tokenizer. Compare with GPT-4 and see ~45% token savings on Turkish text.

TokenizationNLPTurkish

Tokenizer Battle

Soon

Compare tokenization across GPT, Llama, Claude, and more. See why token counts differ.

TokenizationComparison

Attention Heatmap

Soon

Visualize where transformer models look when processing text. See attention patterns.

AttentionTransformersVisualization

Token Probabilities

Soon

See the top candidate tokens at each generation step. Understand how LLMs make choices.

LLMProbabilities

Temperature Playground

Live

Same prompt, different temperatures. See how randomness affects model output.

LLMSampling

RuseN-Gram

Soon

My personal N-Gram model. Old school meets new school language modeling.

N-GramClassic ML