Embeddings

Embeddings convert text into arrays of numbers (vectors) where similar meanings produce similar vectors. This is the technology behind semantic search, recommendations, and RAG.

One Line of Code

embedding = RubyLLM.embed("Ruby is a programmer's best friend")

embedding.vectors        # => [0.0231, -0.0142, ...] (1536 floats)
embedding.model          # => "text-embedding-3-small"
embedding.input_tokens   # => 8

Why Embeddings Matter

Traditional search matches keywords. Embedding search matches meaning:

Your Task

Open embeddings.rb. Three sample texts and a cosine similarity helper are provided. Your job:

1. Generate embeddings for each text with RubyLLM.embed
2. Extract vectors from the results
3. Compare pairs using the cosine_similarity helper
4. Print the scores to see which texts are most similar

vectors = texts.map do |text|
  embedding = RubyLLM.embed(text)
  embedding.vectors
end

# Compare pairs
score = cosine_similarity(vectors[0], vectors[1])
puts "Ruby vs Python: #{score}"

Run it:

$ ruby embeddings.rb

You should see that the two programming texts have a higher similarity score than either compared to the weather text.

Batch Embeddings

Embed multiple texts in a single API call:

result = RubyLLM.embed(["Ruby", "Python", "JavaScript"])

result.vectors.length  # => 3
result.vectors[0]      # => vector for "Ruby"

Choosing Models

RubyLLM.embed("text", model: "text-embedding-3-large")   # higher dimensions
RubyLLM.embed("text", model: "text-embedding-004")       # Google
RubyLLM.embed("text", dimensions: 512)                   # reduced dimensions

Real-World Use Cases

• Semantic search — find documents by meaning, not keywords
• Recommendations — similar content or user preferences
• RAG — retrieve relevant context for AI prompts
• Clustering — group similar documents automatically
• Deduplication — find near-duplicate content