Vector Hyperbolic

Embeds hierarchical data in the Poincaré ball model using ruvector@0.2.25, though you'll need to handle the hyperbolic projection yourself since that version lacks a native Poincaré flag. You generate standard ONNX embeddings first, then normalize them into the unit ball and calculate geodesic distances in your own code. The payoff is logarithmic distance scaling that captures tree depth efficiently, useful for dependency graphs, class hierarchies, or any nested structure where "closer to the root" versus "deep in a subtree" matters semantically. Worth the extra projection step if you're mapping codebases or taxonomies where Euclidean embeddings waste dimensions on hierarchy.