[Feature]: Fuzzy Keyword Search (RapidFuzz) to Enhance BM25 in HelixDB #781
Description
What do you want?
Add fuzzy keyword search support to HelixDB using libraries such as RapidFuzz (or similar), to complement and enhance the existing BM25-based keyword search.
This would allow HelixDB to handle:
Typos and misspellings
Partial matches
Slightly different word forms and variations
The fuzzy matching layer could be applied as a pre-filter or post-ranking step alongside BM25, improving recall while preserving the relevance scoring and performance benefits of the current search system.
Adding fuzzy search aligns strongly with HelixDB’s goal of “Ultra-Low Latency.”
By using highly optimized string-matching algorithms (as provided by RapidFuzz), HelixDB can improve search quality with minimal additional overhead, especially for short queries and user-generated input.
Feature Area
Other
Additional context
Implementation suggestion
Use RapidFuzz (Rust bindings or native implementation) for fast Levenshtein / token-based similarity scoring.
Combine fuzzy scores with BM25 scores for hybrid ranking.
Make fuzzy matching configurable (thresholds, enabled/disabled per query).