archetype

Embeddings

also: vector embeddings · vector database · semantic search

Numeric vectors that represent the meaning of text, images, or signals so similar items sit close together; the basis of semantic search and RAG.

Word embedding illustration: related word pairs such as France and Paris sit in parallel positions in vector space.
Fschwarzentruber / Wikimedia Commons (resized) · CC BY-SA 4.0

An embedding model converts an item (a paragraph from a manual, a part description, an image, a window of sensor data) into a list of numbers. Items with similar meaning land near each other, so you can search by meaning ("motor overheats at startup") rather than exact keywords, cluster similar field reports or find near-duplicate parts. Vector databases, or vector extensions to ordinary databases, store and search them efficiently.

Quality depends on how well the embedding model fits your domain and how content is chunked. Technical documents full of part numbers, units, and tables often need hybrid search that combines embeddings with keyword matching.

Founder takeaway

Test retrieval quality on real queries from your domain, and combine semantic and keyword search for part numbers and specifications.

Ask an engineer