AnniiGCP / VECTORFORGE-AI

What it does

• Three Search Algorithms from Scratch. Hierarchical Navigable Small World (HNSW), KD-Tree with exact metric pruning bounds, and exhaustive Brute Force.
• Three Distance Metrics. Cosine Distance ($1 - \cos\theta$), Euclidean Distance ($L_2$), and Manhattan Distance ($L_1$).
• Local RAG Pipeline. Multi-stage pipeline with sliding-window chunking, Ollama embedding generation, HNSW retrieval, strict source grounding, and SSE token streaming.
• Mathematical PCA Projection. Server-side covariance matrix computation and cyclic Jacobi eigensolver for real-time 2-D visualization.
• Synthetic Benchmarking Suite. Automated benchmarking harness comparing search latency, recall@K, build times, and traversed nodes across dataset sizes up to hundreds of thousands of vectors.