liveBook · Manning

Vector database

As explained in LLM Customization and Fine-Tuning

An operational database layer that provides vector search plus persistence, filtering, replication, multi-tenancy, and observability.

A vector database wraps a vector index with an operational layer: query APIs, metadata filters, persistence, replication, multi-tenancy, and observability. section 4.5.3

Choose the layer to match corpus size and production needs:

  • Use a NumPy array for prototypes with tens of thousands of chunks, an in-process index library such as FAISS for hundreds of thousands, and a vector database for multi-tenant production with millions and audit logging. section 4.5.3
  • Check embedding-version consistency, index-update latency, and filtering performance before choosing; changing the embedder requires re-embedding the corpus. section 4.5.3

Production choices

Pinecone is managed; Weaviate and Qdrant are dedicated vector databases; Postgres with pgvector adds vector-index support to a traditional database. section 4.5.3

How LLM Customization and Fine-Tuning covers it

How deep it goes
Distinguishes database capabilities and scale choices, then flags embedding drift, index-update latency, and filtering performance.
Examples
Names Pinecone, Weaviate, Qdrant, Postgres with pgvector, FAISS, and a NumPy array as scale-specific options.
What you can do after
Evaluate database options against corpus scale, tenancy, audit logging, update frequency, and filtering needs.
How it connects
Connects vector databases to vector indexes, embeddings, approximate nearest-neighbor search, and metadata filters.

Also explained in

Compare how 34 books explain vector database →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of LLM Customization and Fine-Tuning. The section links point to the parts of the book it draws on.