<< back to Guides

AI Core Concepts (Part 17): Embeddings

Embeddings are vector representations of data (text, images, audio, etc.) that capture semantic meaning in a machine-readable format. They're foundational to many AI tasks like search, recommendation, and similarity comparison.

1. What Are Embeddings?

An embedding maps complex input (e.g., a sentence or image) into a dense vector of real numbers in a high-dimensional space.

For example:

2. Embedding Spaces

Embeddings exist in high-dimensional vector spaces (e.g., 512 or 1536 dimensions), where:

from sklearn.metrics.pairwise import cosine_similarity

# Vectors (pretend these came from a model)
vec_cat = [0.2, 0.8, 0.1]
vec_kitten = [0.25, 0.75, 0.15]

similarity = cosine_similarity([vec_cat], [vec_kitten])
print(similarity)

3. Text Embeddings with OpenAI API

import openai

response = openai.Embedding.create(
    input=["artificial intelligence", "machine learning"],
    model="text-embedding-3-small"
)

embedding_1 = response["data"][0]["embedding"]
embedding_2 = response["data"][1]["embedding"]

4. Visualizing Embeddings (Optional)

Use t-SNE, UMAP, or PCA to project high-dimensional embeddings into 2D/3D:

from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

X = [embedding_1, embedding_2, ...]
labels = ["AI", "ML", "Pizza"]

X_proj = TSNE(n_components=2).fit_transform(X)
plt.scatter(X_proj[:, 0], X_proj[:, 1])
for i, label in enumerate(labels):
    plt.annotate(label, (X_proj[i, 0], X_proj[i, 1]))
plt.show()

5. Use Cases of Embeddings

Task Description
Semantic Search Find similar docs/questions
Recommendations Match users and content based on behavior
Clustering/Topic Modeling Group similar items without supervision
Text Classification Use embeddings as features for ML models
RAG (Retrieval-Augmented Generation) Combine vector search with LLMs

6. Embedding Models (Text/Image/Audio)

Type Examples
Text text-embedding-3-small, all-MiniLM-L6
Image CLIP, BLIP-2
Audio Wav2Vec, Whisper embeddings
Multi Gemini, GPT-4o

7. Example: Sentence Similarity with SentenceTransformers

from sentence_transformers import SentenceTransformer, util

model = SentenceTransformer('all-MiniLM-L6-v2')
sentences = ["The cat sits outside", "A dog barks loudly"]

embeddings = model.encode(sentences)
similarity = util.cos_sim(embeddings[0], embeddings[1])

print("Similarity score:", similarity.item())

8. How Are Embeddings Used in LLM Pipelines?

LLMs often use embeddings in RAG systems, tool use, and memory:

# Pseudocode
query_vector = embed("What is a transformer?")
docs = vector_store.similarity_search(query_vector)
context = combine(docs)
response = llm.generate(prompt=context)

9. Common Vector Databases for Storing Embeddings

These allow fast similarity search across millions of vectors.

📚 Further Resources

<< back to Guides