REDDIT · REDDIT// 10d agoOPENSOURCE RELEASE

EVōC Challenges UMAP-HDBSCAN on Embedding Clustering

EVōC is a new Python library for clustering high-dimensional embedding vectors, built specifically for use cases like CLIP, sentence-transformer, OpenAI, and Cohere embeddings. It reworks the UMAP + HDBSCAN approach around embedding data, with a scikit-learn-style API, multi-granularity cluster layers, hierarchy extraction, and near-duplicate detection. The project is positioned as an early beta, but the pitch is clear: better clustering results with less tuning and much faster runtime, with performance said to be competitive with MiniBatchKMeans.

// ANALYSIS

Strong niche fit, especially for teams that already cluster embeddings and are tired of babysitting UMAP + HDBSCAN.

–The differentiation is practical, not flashy: it targets a real pain point in embedding workflows, where dimensionality and compute cost make generic clustering awkward.
–The scikit-learn-compatible API lowers adoption friction for ML engineers who already have clustering pipelines.
–Multi-layer clustering and hierarchy support make it more interesting than a plain “faster KMeans alternative,” especially for topic modeling and analysis workflows.
–The project still reads as early beta, so the claims should be validated on real datasets before treating it as a drop-in replacement.
–Product Hunt presence does not appear to exist yet, so this is primarily a GitHub/PyPI open-source launch rather than a broader consumer product launch.

// TAGS

clusteringembeddingspythonllmhdbscanumapscikit-learntopic-modeling

DISCOVERED

10d ago

2026-04-01

PUBLISHED

11d ago

2026-04-01

RELEVANCE

8/ 10

AUTHOR

lmcinnes