Schema classes, HNSW+BQ, vectorizers, generative search.

Weaviate — GraphQL Vector Search with Modules

Weaviate is an open-source vector database with a GraphQL API. Data is organised into classes (like tables) with typed properties. Built-in vectorizer modules (text2vec-openai, text2vec-cohere) automatically embed text on import. HNSW with Binary Quantisation (HNSW+BQ) compresses index memory 32x. Generative search (RAG) returns LLM-synthesised answers from retrieved objects. Supports hybrid search, multi-tenancy, and replication.

Interactive Pipeline Flow

UnderstandImplementOptimise

📐 Schema — class + properties

Define schema class: {class:"Document", vectorizer:"text2vec-openai", moduleConfig:{model:"text-embedding-3-small"}, properties:[{name:"content",dataType:["text"]}]}.

Architecture Diagrams

Weaviate generative search: retrieve + generate in one query.

Loading diagram...

Key Design Decisions

🧬Use vectorizer modules to simplify ingestion

text2vec-openai auto-embeds on import — no separate embedding pipeline. Configure moduleConfig per class to control model and which properties to vectorize.

💾Enable Binary Quantisation for memory efficiency

HNSW+BQ compresses 1536-dim float32 vectors (6KB each) to 192 bytes — 32x reduction. Minimal recall loss with rescore enabled. Essential for large-scale deployments.

🤖Generative search = zero-boilerplate RAG

Weaviate's generative module does retrieval + synthesis in one GraphQL query. No application-level prompt assembly needed for standard RAG patterns.

Discussion