Class: RubyLlmMesh::Rag::Embeddings

Inherits:
Object
  • Object
show all
Defined in:
lib/ruby_llm_mesh/rag/embeddings.rb

Overview

Zero-dependency bag-of-words style embeddings for local prototyping. Swap for a real embedding provider in production.

Instance Method Summary collapse

Constructor Details

#initialize(dimensions: 256) ⇒ Embeddings

Returns a new instance of Embeddings.



8
9
10
 
# File 'lib/ruby_llm_mesh/rag/embeddings.rb', line 8

def initialize(dimensions: 256)
  @dimensions = dimensions
end

Instance Method Details

#cosine_similarity(a, b) ⇒ Object

Raises:

  • (ArgumentError) 


26
27
28
29
30
31
32
 
# File 'lib/ruby_llm_mesh/rag/embeddings.rb', line 26

def cosine_similarity(a, b)
  raise ArgumentError, "vectors must match length" unless a.length == b.length

  dot = 0.0
  a.each_index { |i| dot += a[i] * b[i] }
  dot
end

#embed(text) ⇒ Object 



12
13
14
15
16
17
18
19
20
 
# File 'lib/ruby_llm_mesh/rag/embeddings.rb', line 12

def embed(text)
  vector = Array.new(@dimensions, 0.0)
  tokenize(text).each do |token|
    index = stable_hash(token) % @dimensions
    vector[index] += 1.0
  end
  normalize!(vector)
  vector
end

#embed_many(texts) ⇒ Object 



22
23
24
 
# File 'lib/ruby_llm_mesh/rag/embeddings.rb', line 22

def embed_many(texts)
  Array(texts).map { |t| embed(t) }
end

#top_k(query, documents, k: 3) ⇒ Object 



34
35
36
37
38
39
40
41
42
 
# File 'lib/ruby_llm_mesh/rag/embeddings.rb', line 34

def top_k(query, documents, k: 3)
  query_vec = embed(query)
  scored = documents.map do |doc|
    text = doc.is_a?(Hash) ? doc[:text] || doc["text"] : doc.to_s
    score = cosine_similarity(query_vec, embed(text))
    { document: doc, score: score }
  end
  scored.sort_by { |row| -row[:score] }.first(k)
end