How do I choose between different distance metrics?

The skill provides a guide: Cosine is best for normalized embeddings, Euclidean (L2) for raw distances, and Dot Product when magnitude is a critical factor.

What indexing methods are covered?

The skill includes patterns for Flat, HNSW (graph-based), and IVF+PQ (quantized) indexing to help you balance search speed versus accuracy.

Does it support hybrid search?

Yes, it includes patterns for combining semantic vector search with traditional keyword filtering and metadata constraints for more precise results.

What is similarity search used for?

It is primarily used for semantic search, recommendation systems, and RAG, where finding similar context matters more than exact keyword matching.

Which vector databases does this skill support?

It provides ready-to-use implementation patterns for Pinecone (managed), Qdrant (high-performance), and PostgreSQL with the pgvector extension.

Vector Similarity Search Patterns

Name: Vector Similarity Search Patterns
Author: drgaciw

bydrgaciw

データサイエンスとML

Implement high-performance similarity search and vector retrieval patterns across multiple database providers.

概要

This skill provides standardized patterns for implementing efficient similarity search within production systems, focusing on vector databases and Retrieval-Augmented Generation (RAG). It equips developers with optimized implementations for popular vector stores like Pinecone, Qdrant, and pgvector, while offering deep insights into distance metrics and indexing strategies like HNSW and IVF+PQ. Whether you're building a semantic search engine, a recommendation system, or scaling an AI application to millions of vectors, this skill ensures best practices for low-latency retrieval, high recall, and advanced reranking workflows.

主な機能

Comprehensive distance metric selection guide (Cosine, L2, Dot Product)
Advanced reranking and cross-encoder implementation patterns
Optimized HNSW and IVF+PQ indexing configurations
Hybrid search and complex metadata filtering workflows
Multi-provider support for Pinecone, Qdrant, and pgvector
0 GitHub stars

ユースケース

Developing high-scale semantic search and recommendation engines
Optimizing vector retrieval latency and recall for production AI apps
Building robust Retrieval-Augmented Generation (RAG) pipelines

概要

主な機能

Comprehensive distance metric selection guide (Cosine, L2, Dot Product)
Advanced reranking and cross-encoder implementation patterns
Optimized HNSW and IVF+PQ indexing configurations
Hybrid search and complex metadata filtering workflows
Multi-provider support for Pinecone, Qdrant, and pgvector
0 GitHub stars

ユースケース

Developing high-scale semantic search and recommendation engines
Optimizing vector retrieval latency and recall for production AI apps
Building robust Retrieval-Augmented Generation (RAG) pipelines