📐 AI Vector Similarity Matrix

RAG & vector database embedding distance engine. Computes Cosine Similarity, Euclidean $L_2$, Dot Product, and Angular separation across semantic embeddings.

Semantic Embedding Scenario:

Vector A (Query Embedding)

Vector B (Chunk Embedding)

COSINE SIMILARITY

+0.994

Top-1 RAG Match

ANGULAR DISTANCE

6.2°

Arc Separation

EUCLIDEAN (L2)

0.082

Geometric Distance

Production RAG Threshold: In vector databases (Pinecone, Qdrant, pgvector), a Cosine Similarity score > 0.82 generally indicates strong semantic relevance for context retrieval, while scores < 0.65 frequently introduce hallucinated noise into the model's generation loop.