Show HN: Clark Hash, 32x smaller searchable sketches for embeddings

Category: library

Tags: embedding-compression, vector-search, rust-library

Score: 7.3/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 7)

Clark Hash implements a stateless sparse Johnson-Lindenstrauss projection with fixed scalar quantization to produce compact, searchable sketches of neural embeddings, achieving 32x storage reduction for vector databases without requiring training or codebook calibration. It combines embedding compression with asymmetric query scoring in Rust, making it interesting for online semantic memory, retrieval prefilters, and edge deployments where memory and bandwidth are constrained.

Target audience: backend devs, data engineers, AI/ML engineers

Repository: https://github.com/clark-labs-inc/clark-hash · Rust · NOASSERTION · 1 stars

View on Hacker News