use case · practical
Keyword search fails when the words don't match but the meaning does. Here, every line of your corpus becomes a vector once; your query becomes a vector too; and results rank by cosine similarity. Ask "how do I get my money back?" and a line about refunds wins even though it never says "money back."
The corpus is embedded once and reused for every query — that's the whole trick behind a fast vector index. In a real app you'd store those vectors (IndexedDB, a vector DB) so you only embed new content. Everything here stays in the tab.
← Back to MiniLM · Next: Wild →