Atlas Hybrid Search
A Next.js app for testing MongoDB Atlas hybrid search, combining full-text and vector search queries through a unified $search stage with adjustable rank-fusion parameters.
Atlas Hybrid Search is a Next.js application for experimenting with full-text search and vector search against a MongoDB Atlas instance using its unified $search aggregation stage. It runs text search through the compound operator, vector search through the vectorSearch operator, and hybrid search that merges both using $rankFusion or $scoreFusion, plus prefiltered vector search that applies lexical criteria before the vector search runs.
Sliders in the UI let a user adjust the hybrid search algorithm's parameters, and the results panel shows the weighted and normalized text and vector scores that were summed into each document's final score, so a user can see how much each search type contributed. Lexical prefiltering can reduce the candidate set for vector search to documents that already match a text query, using a consolidated search index with vector fields. Vector queries are embedded using OpenAI's ada-002 model by default.
The app supports MongoDB Atlas 7.0.2 and newer, and is MIT licensed.