HNSW Visual Explorer

This visualisation uses AI Generated code, finetuned for the best visualisation, not code quality

hnsw.mov

Interactive Raylib explorer for understanding hierarchical navigable small-world graphs: how the structure is built, how layer descent works, and why greedy graph routing can find strong ANN candidates quickly.

What This Visualisation Shows

• Hierarchical layer construction and neighbor linking
• Query descent from sparse upper layers to dense lower layers
• Greedy graph routing and candidate refinement
• 2D and 3D views of the graph, embeddings, and search state

Visual Map

flowchart LR
    A["Insert Points"]
    B["Assign Random Levels"]
    C["Connect Neighbors"]
    D["Upper-Layer Routing"]
    E["Lower-Layer Refinement"]
    F["Nearest Candidates"]

    A --> B
    B --> C
    C --> D
    D --> E
    E --> F

Loading

Controls

• q: quit
• space: pause or resume the current build/search animation
• r: reset the current scene
• left drag, mouse wheel, and right-click interactions are used on the embedding views as labeled in-app

Run

make run