# Embeddings

Embeddings are numerical representations of text, or other data, that capture its meaning as a list of numbers, so that concepts with similar meaning end up mathematically close to each other.

They're the underlying technology that makes semantic search possible, turning language into something a computer can compare directly.

**Embeddings** — Embeddings are numerical vector representations of text, or other data like images, that capture semantic meaning. An AI model converts a piece of text, a resume, a job description, a search query, into a list of numbers positioned in a high-dimensional space, such that texts with similar meaning end up located close together in that space, even if they don't share any of the same words. Comparing two embeddings, mathematically measuring the distance between their vectors, is how a system determines how semantically similar two pieces of text actually are. This is a fundamentally different comparison than keyword matching, since two texts can be very close in embedding space while sharing almost no literal vocabulary, as long as they're talking about the same underlying concept.

Embeddings sit underneath semantic search, recommendation systems, and any tool that needs to compare meaning rather than exact text. Once resumes, job descriptions, or candidate profiles are converted into embeddings and stored, a search system can find the closest matches to a new query almost instantly, since it's comparing vectors mathematically rather than scanning full text for each search. The quality of an embedding model, how well it captures true meaning rather than surface patterns, directly determines how good the resulting search or matching actually is.

Intervieux stores candidate and query embeddings in pgvector, which is what powers the semantic talent search employers use to find candidates by meaning, comparing an employer's plain-language query against candidate data by vector proximity rather than exact keyword matches.

## Frequently asked questions

### Do embeddings only work with text?

No, though text is the most common use in hiring. Embeddings can also represent images, audio, or other data types, converting any of them into a comparable numerical form.

### How are embeddings different from a keyword index?

A keyword index matches exact words. Embeddings capture meaning as numbers, so two texts about the same concept can be found as similar even when they don't share any literal words in common.

### Are embeddings something a candidate or recruiter interacts with directly?

No, they're a behind-the-scenes technical layer. Users interact with the search or matching results embeddings enable, such as a plain-language talent search, without ever seeing the underlying vectors.

## Related pages

- [Semantic search](/glossary/semantic-search)
- [Job matching algorithms](/glossary/job-matching-algorithm)
- [Browse open roles](/jobs)

## Get found by meaning, not just keywords

Intervieux's semantic talent search compares candidates by embeddings, so a strong fit can surface even when a resume doesn't use an employer's exact search terms.

Start practicing free: https://www.intervieux.ai/register · Hire with Intervieux: https://www.intervieux.ai/employers/signup