Vector Databases

Why Vector Databases?

Traditional databases search by exact match or keyword. Vector databases enable semantic search — finding content by meaning rather than exact words. They use Approximate Nearest Neighbor (ANN) algorithms for fast similarity search across millions of vectors.

Popular Vector Databases

Key Concepts

• Indexing algorithms: HNSW (most common), IVF, PQ — trade-offs between speed, memory, and accuracy
• Metadata filtering: Combine vector similarity with traditional filters (date, category, user)
• Hybrid search: Combine dense vectors (semantic) with sparse vectors (keyword/BM25) for better results
• Namespaces/collections: Organize vectors by tenant, document type, or use case

Example: pgvector with PostgreSQL

CREATE EXTENSION vector;

CREATE TABLE documents (
    id SERIAL PRIMARY KEY,
    content TEXT,
    embedding vector(1536),
    metadata JSONB
);

-- Find 5 most similar documents
SELECT content, 1 - (embedding <=> query_embedding) AS similarity
FROM documents
ORDER BY embedding <=> query_embedding
LIMIT 5;

🌼 Daisy+ in Action: Vector-Ready Architecture

The Daisy+ architecture includes Redis for caching and could extend to vector stores (like pgvector in PostgreSQL 15) for semantic retrieval across ERP records — enabling questions like "find products similar to this description" or "what past projects match this RFP?" Since Daisy+ already runs PostgreSQL, adding pgvector is a natural extension of the existing infrastructure.