Back to VectorScale
Engineering & Systems Research

VectorScale Engineering Blog

Deep dives on Rust storage engines, binary quantization benchmarks, and multi-tenant RAG architecture.

Architecture9 min read

Inside VectorScale v2.4: Building a Zero-Copy Hybrid HNSW + BM25 Engine in Rust

How we unified dense graph traversal and sparse inverted indexes over memory-mapped NVMe pages to eliminate cross-service network hops in production RAG pipelines.

KT
Dr. Kenji Takahashi, Chief Architect
May 14, 2025
Benchmarks7 min read

Benchmarking 32x Binary Quantization with Asymmetric Oversampling on 100M Vectors

Detailed recall@10 and p99 latency measurements across OpenAI, Cohere, and Voyage embeddings when pairing 1-bit quantization with disk-backed float16 rescoring.

SA
Sofia Alvarez, Principal Storage Engineer
April 28, 2025
Best Practices6 min read

Designing Multi-Tenant RAG Without Index Sprawl or Noisy Neighbors

Why spinning up one cluster per enterprise customer breaks down past 1,000 tenants—and how partitioned vector segments solve isolation at 10x lower cost.

LO
Liam O'Connor, Solutions Engineering
April 11, 2025
Systems Dispatch

Get VectorScale benchmarks and engine notes monthly

No marketing fluff—only storage engine internals, recall benchmarks, and release advisories.