What is it?
An embedding is a numerical representation that captures useful meaning or features.
A model maps items such as text, images, or products into a vector. Similar items tend to have nearby vectors in the chosen representation space.
05 · Embeddings, RAG & Vector Search · Topic 01
What are Embeddings?
An embedding represents text or other data as numbers that a computer can compare.
Think of it like...
Think of it like placing ideas on a map.
Items with similar meaning are placed near each other. The map helps a system compare relationships, but it is not a literal map of human meaning.
How it works
1 · InputReceive the itemRead text, an image, or another data object.
2 · EncodeApply an embedding modelConvert the item into a vector.
3 · StoreSave the vectorKeep it with an ID and source metadata.
4 · CompareMeasure similarityCompare a query vector with stored vectors.
5 · UseRetrieve or recommendUse nearby items in a downstream task.
Real-world examples
Everyday example · Search
Input: a question about changing a password
AI / System: compares its embedding with help articles
Output: the most relevant articles
Business example · Product matching
Input: a product description
AI / System: compares product embeddings
Output: similar products for review or recommendation
What it is NOT
Embedding ≠ Keyword
Embedding: represents patterns and relationships as numbers.
Keyword: is a literal word or phrase match.
Embedding ≠ Database
Embedding: is a vector representation.
Database: stores vectors, metadata, or other records.
Embedding ≠ Meaning Itself
Embedding: is a model-generated representation.
Meaning itself: cannot be guaranteed by a vector alone.
Related concepts
Data → Embedding Model → Vector → Vector Search → Retrieved Items
Explore next
Remember this
An embedding turns data into numbers that a system can compare.
Independent explainer