Weaviate

Open-source vector database company providing semantic and hybrid search infrastructure. Competes with Pinecone, Qdrant, Milvus.

Notable Contributions

Cross-encoder reranking — Weaviate has documented and popularized the bi-encoder + cross-encoder two-stage pipeline for semantic search:

• Stage 1: Bi-encoder ANN (fast, pre-computed embeddings)
• Stage 2: Cross-encoder reranker (accurate, query-aware, applied to small candidate set)

Key insight: Bi-encoders are fast because document embeddings are computed offline; cross-encoders are accurate because query and document are encoded together but can’t scale to full corpora.

Weaviate supports cross-encoder reranking as a native feature in its search pipeline.

Block-Max WAND — Weaviate implemented Block-Max WAND in v1.29 (technical preview), achieving 5–10x p50 keyword search speedups and 50–90% index size reduction vs. their prior WAND baseline. Implemented by André Mourão and Joon-Pil (JP) Hwang.

Key Articles

• Using Cross-Encoders as Reranker in Multistage Vector Search

• BlockMax WAND - How Weaviate Achieved 10x Faster Keyword Search

Tools

• Weaviate Vector DB

People

• Laura Ham
• André Mourão — search engineer; Block-Max WAND implementation
• Joon-Pil (JP) Hwang — search engineer; Block-Max WAND implementation

Concepts

Block-Max WAND · WAND · Hybrid Search · Cross-Encoder