Crate sevensense_vector

Crate sevensense_vector 

Source
Expand description

§sevensense-vector

Vector database operations and HNSW indexing for the 7sense bioacoustics platform.

This crate provides:

  • Local HNSW index with 150x search speedup over brute-force
  • Optional Qdrant client wrapper for distributed deployments
  • Collection management
  • Similarity search with filtering
  • Batch operations and persistence
  • Hyperbolic embeddings for hierarchical relationships

§Architecture

Following Domain-Driven Design:

sevensense-vector
├── domain/              # Core entities, value objects, repository traits
│   ├── entities.rs      # EmbeddingId, HnswConfig, SimilarityEdge
│   ├── repository.rs    # VectorIndexRepository, GraphEdgeRepository
│   └── error.rs         # VectorError
├── application/         # Service layer with use cases
│   └── services.rs      # VectorSpaceService
└── infrastructure/      # HNSW implementation and storage adapters
    ├── hnsw_index.rs    # Local HNSW index
    └── graph_store.rs   # Edge storage

§Performance Targets

  • 150x search speedup over brute-force linear scan
  • Sub-millisecond queries for up to 1M vectors
  • Efficient batch insertion with parallelization

§Example

ⓘ
use sevensense_vector::prelude::*;

// Create a vector space service
let config = HnswConfig::for_dimension(1536);
let service = VectorSpaceService::new(config);

// Add embeddings
let id = EmbeddingId::new();
let vector = vec![0.1; 1536];
service.add_embedding(id, vector).await?;

// Search for neighbors
let query = vec![0.15; 1536];
let neighbors = service.find_neighbors(&query, 10).await?;

Re-exports§

pub use domain::entities::EmbeddingId;
pub use domain::entities::HnswConfig;
pub use domain::entities::SimilarityEdge;
pub use domain::entities::EdgeType;
pub use domain::entities::VectorIndex;
pub use domain::entities::Timestamp;
pub use domain::repository::VectorIndexRepository;
pub use domain::repository::GraphEdgeRepository;
pub use application::services::VectorSpaceService;
pub use application::services::Neighbor;
pub use application::services::SearchOptions;
pub use infrastructure::hnsw_index::HnswIndex;

Modules§

application
Application layer for the Vector Space bounded context.
domain
Domain layer for the Vector Space bounded context.
error
Error types for vector operations
infrastructure
Infrastructure layer for the Vector Space bounded context.
prelude
Prelude module for convenient imports Common imports for vector operations.

Constants§

VERSION
Crate version information

Functions§

cosine_distance
Compute the cosine distance between two vectors.
cosine_similarity
Compute the cosine similarity between two vectors.
euclidean_distance
Compute the Euclidean (L2) distance between two vectors.
exp_map
Exponential map: project from tangent space at origin to the Poincare ball.
log_map
Logarithmic map: project from Poincare ball to tangent space at origin.
mobius_add
Mobius addition (gyrovector addition).
normalize_vector
Normalize a vector to unit length (L2 normalization).
poincare_distance
Compute the Poincare distance between two points in the Poincare ball.