ConstellaVS Vector Search Engine
ConstellaVS is a vector similarity search engine indexed by Navigable Small World (NSW) graphs, capable of returning approximate k-nearest neighbors of an arbitrary query vector. The executable processes commands on the standard input and prints results to the standard output, allowing it to be used interactively in the terminal or programmatically as part of a larger pipeline.
This demo represents a motivational use case for ConstellaVS. The following program is a simple script running ConstellaVS in the background, loaded with 100,000 vectors from the glove-wiki-gigaword-50 dataset where each vector is an embedding of a frequent word from the English language. The script allows the user to enter a word and how many linguistically-closest words to get, and - if the input word is in the dataset - it will return the most similar words from the dataset based on a query executed by ConstellaVS.
Enter a single English word into the query box and specify the number of neighbors (k). The demo restricts queries to an upper bound of k=200.