Embeddings are the mathematical foundation of modern AI applications for semantic similarity. Text, images or other data are translated into high-dimensional vectors – typically 768 to 4096 dimensions. Content with similar meaning lies close together in that space, even when different words are used.
In recruiting, embeddings underpin semantic search and job matching. A job posting for "social worker" and the CV of a "social pedagogue" have similar embeddings even though the terms differ. A well-designed system thus surfaces roles that would remain hidden under classic keyword filters.
Technically, embeddings are produced either with general-purpose models (OpenAI, Cohere, Voyage) or with bespoke domain-specific fine-tunes. They are stored in vector databases that enable similarity search within milliseconds.
Lunigi uses embeddings to continuously compare profiles against current openings – the results land compactly in the daily email digest.