Back to Home
Retrieval-Augmented Generation (RAG)

Embeddings in RAG

Embeddings in RAG turn document chunks into vectors so the system can search by meaning.

This demo uses simple learning vectors, not real AI embeddings.

Interactive Playground

Document Chunks

Live Visualization
📄 Chunks → 🔢 Embeddings → 🔍 Compare with Query → 🎯 Best Match

Query Vector

—

[0, 0, 0, 0, 0, 0]

Chunks Ranked by Similarity

Statistics

4
Chunks
0
Query Tokens
0
Comparisons Made
—
Best Matching Chunk
0%
Highest Similarity

How It Works

📄
Document
Chunks
→
🔢
Embeddings
→
❓
Query
Embedding
→
🔍
Similarity
Search
→
🎯
Relevant
Chunk
Similar meaning, similar vectors. Even without shared words, "money back" and "refund" can point in a similar direction.
Cosine similarity measures direction, not length. It checks how closely two vectors point the same way.
Real embedding models use hundreds or thousands of dimensions learned from massive text datasets, this demo uses just a handful to stay easy to visualize.
💡
Key Takeaway

RAG uses embeddings so it can find useful information by meaning, not just exact words.

Previous: Metadata Filtering Next: Vector Search