RAG & Retrieval Lab · Intermediate
Vector Search Playground
Compare cosine, dot-product and Euclidean similarity and run semantic search on a plotted vector space.
3 similarity metrics
Vector space plot
Semantic search
Vector search uses mathematical similarity metrics to find semantically similar documents. Unlike keyword matching,
it understands context and meaning through high-dimensional vector representations of text.
Understanding Similarity Metrics
Cosine Similarity
Measures the angle between vectors, ignoring magnitude. Perfect for text where document length doesn't matter.
Dashed lines show different angles from query point
Range: -1 to 1 (higher = more similar)
Euclidean Distance
Measures straight-line distance between points. Considers both direction and magnitude differences.
Concentric circles show distance boundaries from query
Range: 0 to ∞ (lower distance = more similar)
Dot Product
Combines angle and magnitude. Favors documents that are both similar in direction and have high values.
Grid lines show projection axes for magnitude calculation
Range: -∞ to ∞ (higher = more similar)
2D Visualization Guide
The scatter plot shows a 2D projection of high-dimensional document vectors. Each point represents a document, and the red star is your query. Point sizes change based on similarity scores, and visual patterns help you understand how each metric behaves differently:
- Cosine: Similar documents cluster by direction from origin
- Euclidean: Similar documents are physically close together
- Dot Product: Considers both distance and position in space
Select the mathematical method for calculating similarity between vectors
Vector Search Results
Enter a query to see vector search results
Keyword Search Results
Enter a query to see keyword search results