hybrid-search-rrf

Reciprocal Rank Fusion (RRF) improves hybrid search by aggregating full-text and vector search rankings into a unified score. It reconciles exact keyword matches with semantic similarity, handling overlapping results to produce stronger final rankings.

You can merge full-text and vector search results in Convex by applying Reciprocal Rank Fusion. This approach combines searchIndex and vectorIndex outputs, reconciling exact keyword matches with semantic similarity for improved retrieval quality.

Yes, hybrid search supports adjustable weight configuration for RRF merging. This allows you to tune the influence of full-text versus vector rankings when aggregating overlapping results into the final relevance score.

Hybrid search works with both Convex searchIndex and vectorIndex. It pulls results from both indexes and merges them using Reciprocal Rank Fusion to boost relevance across traditional keyword and semantic vector queries.

An admin toggle allows you to disable hybrid search and fall back to text-only results. This provides a necessary fallback path when vector search is not needed or encounters issues.

Use hybrid search instead of vector search alone when queries benefit from both exact keyword matches and semantic similarity. RRF merging boosts relevance for diverse queries that rely on both precise terms and contextual meaning.