Embeddings and Semantic Search

An embedding maps text into a dense vector so similar meaning sits nearby in space—powering semantic search and RAG retrieval.

Similarity search

import math

def cosine(a, b):
    dot = sum(x * y for x, y in zip(a, b))
    na = math.sqrt(sum(x * x for x in a))
    nb = math.sqrt(sum(y * y for y in b))
    return dot / (na * nb + 1e-9)

# Higher cosine ≈ more similar meaning (illustrative 3-D vectors)

Embedding models

Dedicated embedding APIs (e.g. small transformer encoders) produce vectors for passages and queries. Use the same model for indexing and querying.

Not magic

• Synonyms and paraphrases match well; exact SKU codes may need hybrid keyword + vector search
• Stale docs in the index produce confident wrong answers—operate the corpus

Important interview questions and answers

1. Q: Cosine similarity vs Euclidean?
   A: Cosine is standard for normalized embedding vectors; scale-invariant direction match.

Self-check

1. What is an embedding used for in RAG?
2. Why use the same model for index and query?

Tip: Re-embed the corpus when you change embedding models—mixed indexes break retrieval.