Search by Meaning, Not Keywords

embeddings and semantic search

Part of: Semantic Search Engine

Over the next eight lessons you'll build a search tool that finds results by meaning , not by matching keywords. Search "puppy training tips" and it should still surface a document titled "how to teach your dog to sit," even though the two share almost no words. By the end you'll have a working engine you can point at any small dataset and query in plain English. Keyword search hits a wall Classic search (grep, SQL LIKE, a plain search box) matches literal substrings. It can't tell that "car" and "automobile" mean roughly the same thing, or that "return policy" and "how do I send this back" are the same question. It's fast and simple, and it's blind to meaning. What an embedding is An embedding is a list of floating-point numbers, produced by a model, that places the meaning of a piece of text at a point in space. Two texts with similar meaning get embeddings that land close together; two unrelated texts land far apart. A real embedding vector runs from a few hundred to over a thousand numbers long, but the intuition matches a 2D map: closeness in the vector space is closeness in meaning. Calling a real embeddings API Claude's Messages API has no embeddings endpoint. Anthropic's do

Challenge: Same Dimension or Bust