ParqDB – Vector search in the browser from Parquet over HTTP"
- Source
- search.parqdb.io
- Author
- petrizhang
- Date
Context
ParqDB's Wikipedia search demo moves similarity-search computation into the browser. It publishes an index of numerical article representations as Parquet files in object storage. The browser converts a question into the same kind of representation, requests the necessary file segments over HTTP and ranks candidate matches locally. The demo page describes this workflow over 100,000 articles.
The architectural change is that object storage serves file segments while the browser performs the search calculations, removing a separate query server from this design. It still requires network requests to storage. The documented workflow therefore supports a browser-executed search system, without establishing that every aspect of a user's activity is hidden from infrastructure providers.
- embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
Checking sign-in…
Loading comments…


