Embeddings (also called vector embeddings) are the technology that lets AI "understand" the semantic closeness between texts. In practice, every word, sentence or document is turned into a numerical vector: a list of coordinates in a high-dimensional mathematical space (typically 768 to 8,192 dimensions). In that space, texts about the same subject end up close to one another, regardless of the exact words they use.
For the legal field, this technology is a game changer. Classic legal search relies on keywords: if you search for "wrongful dismissal" but the ruling uses "unjustified termination of the employment contract", traditional search fails. With embeddings, semantic search understands that both phrasings refer to the same concept and returns the relevant results. This is the foundation of the RAG systems used by modern legaltech solutions.
Embeddings are stored in specialized vector databases (Qdrant, Pinecone, Weaviate) optimized for similarity search at scale. Indexing millions of court decisions, scholarly commentaries or legislative texts as embeddings enables legal search that is smarter, faster and more comprehensive than traditional approaches.