Semantic search — embeddings for non-technicians

Most search works the way a library index does. You type a word, the system looks for that exact word. Type "meeting" and you find notes containing "meeting". Type "conference" and you get different results — even if they describe the same thing.

What embeddings are

When you save a note in Spote, the text is sent to an AI model that converts it into a numerical representation of the meaning of the text. This is called an embedding.

Two notes about the same topic will produce similar numbers, even if they share no words. A note about "Q3 budget" and a note about "autumn spending plan" will land close to each other. A note about coffee duty will land somewhere else entirely.

How Spote uses this

When you search, your search phrase is converted into an embedding using the same model. Spote then finds notes whose embeddings are close — not notes that contain the same words, but notes that mean something similar.

The same logic powers similar notes. When you open a note, Spote compares its embedding to all your other notes and surfaces the ones that are conceptually related. You might open a quick thought from six months ago and find it next to a more developed note from last week, without ever having linked them.

The practical upside

You do not need to tag perfectly or file things correctly to find them later. Write naturally, save quickly. The structure is already in the meaning.

#semantics #embeddings