LEVEL 01 · LESSON 04
Embeddings
Embeddings make meaning searchable through numerical similarity.
What it is
An embedding is a list of numbers representing features of an input, such as text. For semantic search, the model is trained so useful similarities are reflected in the relationships between these lists.
Why it matters
It lets software retrieve information even when the question and document use different words. This is valuable when customers do not know your internal terminology.
A business example
A customer asks about getting their money back. A search using embeddings may find a policy headed ‘Refund eligibility’ even without an exact phrase match. The policy still needs to be checked for the right product and date.
How it actually works
An embedding model converts documents and queries into vectors. A search system compares these vectors using a similarity measure and returns nearby candidates. There is no universal vector for a sentence: the model determines the representation. Document and query vectors must use compatible embedding spaces. Similarity can miss exact identifiers, negation, or business exceptions, so keyword search and metadata filters often complement it.
A common misconception
An embedding is not a database record of truth, and a high similarity score does not establish that a passage answers the question.
ONE SENTENCE TO REMEMBER
Embeddings make meaning searchable through numerical similarity.
CHECK YOUR UNDERSTANDING
A semantic search returns a passage with a very high similarity score for a customer’s question. What can you reasonably conclude?
Show the answer
Answer: A.
Embeddings place inputs with related meaning close together, so a high score signals relatedness. It does not establish that the passage answers the question, and similarity can miss exact identifiers, negation, or business exceptions. Keyword search and metadata filters often complement it.
Take it into your next meeting
Where do your customers and internal documents use different words for the same thing?
One more concept, understood.
Mark this lesson complete to keep your place.