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

Interactive Raylib lab for understanding the ScaNN retrieval pipeline: coarse partitioning, candidate pruning, anisotropic scoring, survivor rescoring, and full-query execution traces.

• The coarse-to-fine retrieval pipeline behind ScaNN-style search
• How partitions and shortlist survivors reduce full-dataset work
• Where heap cutoffs and candidate scoring drive the final top-k
• A query trace that exposes the major state transitions of one search

flowchart LR
    A["Dataset"]
    B["Coarse Partition"]
    C["Shortlist Survivors"]
    D["Anisotropic Scoring"]
    E["Top-k Heap"]
    F["Final Results"]

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

• q: quit
• Page navigation can be driven by the top chips or the keyboard hints shown in-app
• Query and page-specific interactions are described at the bottom of the active page

make run