Vector Databases

Introduction

A vector database stores embeddings and supports fast nearest-neighbour search. Choices: Pinecone, Weaviate, Qdrant, Chroma, Milvus, pgvector. Pick one that fits your scale and ops.

Beginner analogy: a Google Maps for vectors — find the closest neighbours in milliseconds, even at billions of points.

Understanding the topic

Core concepts:

• Approximate Nearest Neighbour (ANN) indexes: HNSW, IVF, ScaNN.
• Metadata filters: combine vector + structured filters.
• Hybrid search: vector + BM25.
• Sharding for scale; replicas for HA.
• Backups and re-index strategies are critical.

Syntax reference

Visual workflow / architecture:

query → embed → ANN search ──► top-k vectors
                          │
                          ▼
                     fetch original chunks

Real-world use

Pinecone (cloud), Qdrant (self-host), Chroma (dev), pgvector (Postgres add-on), MongoDB Atlas Vector.

Best practices

• Use hybrid search (vector + keyword) for production.
• Plan backups + re-indexing from day one.

Common mistakes

• Choosing a vector DB based on hype — match it to ops capacity.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Three popular vector DBs.
• Q2. Hybrid search — what and why?
• Q3. Scenario: pgvector is fast at 1M rows but slow at 100M. What changes?