Vector Search & LSH Hash Table Playground

Experience how RCRE indexes millions of multimodal items using L2 Spherical Normalization, Locality-Sensitive Hashing (LSH), and Arena-Backed Open-Addressing Hash Tables to deliver instant Top-K search with zero heap allocations.

📚 Corpus Ingestion 50 Items
⚙️ Index Hyperparameters
Embedding Dimensions (D) 64-D (Spherical L2)
LSH Hyperplanes (Bits) 64 Hyperplanes
Multi-Probe Radius (r) Radius = 1 (Neighbors)
r=0 (Exact) r=1 (1-Bit Flips) r=2 (2-Bit Flips)
Top-K Results Limit K = 10
Hash Table Buckets Activity 38 Active
0.12 ms Query Latency
100.0% Recall@K vs Exact
92.0% Corpus Pruned
17 Buckets Probed
0 B Heap Allocation
📐 RCRE Invariant Verified: $\|\mathbf{u} - \mathbf{v}\|_2^2 \equiv 2(1 - \cos\theta) \equiv 2(1 - \langle \hat{\mathbf{u}}, \hat{\mathbf{v}} \rangle)$. Cosine Similarity & Euclidean Distance are mathematically equivalent on the unit hypersphere.
0 LSB Drift
Query 64-bit LSH: 0x0000000000000000 Primary Bucket: #0
# Similarity L2 Dist Hamming LSH Match Matched Text Payload