⬡ Hub
Skip to content

Vector Databases

A Vector Database is specialized infrastructure designed to store, manage, and search massive quantities of high-dimensional vectors (mathematical arrays of numbers). Vector databases are the mandatory backbone of modern AI applications like RAG (Retrieval-Augmented Generation) and Semantic Search.

1. What is a Vector?

As learned in the NLP: Word Embeddings module, AI models (like OpenAI's text-embedding-3-small or Amazon Titan) convert entire documents, images, or audio files into lists of floating-point numbers. For example, the sentence "The quick brown fox" might become a 1536-dimensional array: [0.12, -0.45, 0.88, ... 0.04]. This array mathematically defines the "meaning" of the sentence.

2. Why Relational Databases (SQL) Fail Here

If you store a document in Postgres and a user searches for "automobile", a standard LIKE '%automobile%' SQL query will fail to find documents that use the word "car" or "vehicle". SQL performs exact keyword matching.

Vector Databases perform Similarity Search. When a user searches for "automobile", the query is converted into a Vector. The database mathematically calculates the distance (e.g., Cosine Similarity) between the Search Vector and all Document Vectors. The Vectors for "automobile" and "car" will be mathematically very close to each other, allowing the DB to retrieve the correct document regardless of exact keyword matches.

3. Core Mechanics: ANN (Approximate Nearest Neighbor)

Calculating the exact mathematical distance between a search query and every single one of 100 Million documents would take minutes and crash your server. Vector Databases use algorithms like HNSW (Hierarchical Navigable Small World) to approximate the search. It builds a graph where similar vectors are linked. Instead of scanning everything, the search "hops" across the graph, skipping 99% of irrelevant data, returning results in milliseconds.

4. Industry Standard Vector Databases

  • Pinecone / Weaviate: Fully managed cloud-native vector databases. Heavy enterprise use.
  • ChromaDB / FAISS (by Meta): Lightweight, open-source libraries often run locally or embedded in Python applications. Highly used in LangChain ecosystems.
  • pgvector: An extension for PostgreSQL that allows it to store and search vectors alongside standard relational data.
  • AWS OpenSearch Serverless: Amazon's primary serverless vector engine, heavily integrated with Bedrock Knowledge Bases.

How to execute the examples:

Go to the Examples/ folder and run the script: python GenAI_VectorDB_FAISS.py