Glossary Graph

Concept Graph

Every glossary term is a dot. In the semantic layout, terms sit near each other when their definitions mean similar things: each definition was embedded with the same MiniLM sentence model the RAG lab runs in your browser, then projected to 2D with UMAP. Colours are k-means clusters over those embeddings; lines are the entries’ own “related terms” links, and bigger dots have more of them.

Switch to link structure to see the same graph laid out by its links alone. Where the two pictures disagree, terms are worded alike but not cross-referenced, or linked across topics. The 3D view places terms by a 3D projection of the same embeddings; drag to orbit around it.

Relationships (click to show or hide)

Arrows point from prerequisite to the term that builds on it, from a specific term to its general one, from a part to its whole, and from a user to its tool. Hover or tap a line for its relationship.

Clusters (k-means on the embeddings, named by their most-linked term)

  • k-nearest neighbors · 6
  • Vanishing gradient · 14
  • Large language model · 12
  • Transformer · 12
  • Cross-entropy · 17
  • Regularization · 21
  • Variance · 16
  • Generative model · 11
  • Semantic search · 14
  • Gradient descent · 16
  • Cosine similarity · 13