VectorCore: Low-Level Vector Search & HNSW Indexing Engine

VectorCore is a lightweight, zero-dependency vector search engine built from scratch in Python and NumPy. It implements SIMD-friendly vector distance metrics, an exact brute-force baseline index, and a Hierarchical Navigable Small World (HNSW) graph index with binary disk serialization.

Key Features

• Vectorized Metric Kernels: Optimized Euclidean (L2) and Cosine distance implementations.
• HNSW Graph Index: Approximate Nearest Neighbor (ANN) search using greedy graph traversal with configurable ef_construction and ef_search beam width.
• Exact Flat Index: Linear-scan baseline providing 100% ground-truth recall validation.
• Binary Serialization: Zero-copy disk persistence protocol (.vcore) preserving index topologies and high-dimensional vector embeddings.

Benchmark Results

Evaluated on 5,000 vectors (128 dimensions) queried with 100 randomized vectors at $k=10$:

• Performance Gain: 2.44x faster search latency over brute-force linear scanning.

Quickstart

1. Installation

git clone [https://github.com/irtazirfan08-source/VectorCore.git](https://github.com/irtazirfan08-source/VectorCore.git)
cd VectorCore
pip install -r requirements.txt