Embedding

An embedding is a dense numerical vector representation of a piece of content — text, image, audio, or structured record — that encodes semantic meaning in a way amenable to similarity computation, clustering, and downstream machine learning. In modern AI stacks, text embeddings come from purpose-built models like OpenAI's text-embedding-3-small (1536 dimensions) and text-embedding-3-large (3072 dimensions), Cohere's embed-english-v3.0 and embed-multilingual-v3.0, Google's text-embedding-004, Voyage AI's voyage-3, and open-weight options like BAAI bge-large, Nomic-embed-text, and Jina embeddings. The metric of similarity is almost always cosine similarity (or its equivalent, dot product for unit-normalized vectors) — two passages with similar meaning produce vectors with high cosine similarity. To use embeddings practically: pass each document through the embedding API (or run a local model with Sentence Transformers in Python), store the resulting vectors in a vector database keyed by document ID, then at query time embed the user's question with the same model and run a dense retrieval search. Picking the wrong embedding model is the most common RAG failure — domain-mismatched embeddings produce irrelevant retrieval. AI governance teams track which embedding model produced which index because re-embedding a billion documents is expensive and disruptive; embedding model versioning belongs in the model registry alongside the LLM itself.

Embeddings as the latest layer on 25 years of data normalization: Centralpoint generates embeddings on-premise using open-weight models like Llama, Qwen, and Nomic — never sending content to a third-party embedding API — and stores them in the same governance envelope that has protected client data for 25 years. The embedding pipeline reuses the dedup, aggregation, and sensitivity-filtering logic Oxcyon built over a quarter-century, with chatbots deploying through one line of JavaScript.


Related Keywords:
Embedding,Embedding,Oxcyon, AI, AI Governance, Generative AI, Inference, Inference, Inferencing, RAG, Prompts, Skills Manager,