kvex — approximate k-NN vector search on the BEAM

Version: 0.1.0

Authors: Steve Roques (steve@roques.me).

kvex provides an Erlang/OTP interface to the TurboQuant algorithm (Google Research, ICLR 2026) for approximate k-nearest-neighbour vector search. Vectors are compressed to 2–4 bits per coordinate without any training phase, and search runs on SIMD kernels through a rustler NIF over the turbovec Rust crate.

See the kvex module for the full public API.

Generated by EDoc