RapidFuzz supplied normalized fuzzy scoring for part, category, supplier, and description fields. Its documented extraction and weighted-ratio APIs fit the small in-process catalog well, and representative misspellings passed after ranking and cutoff logic were tuned.
- What worked
- The library enabled typo-tolerant ranked results without a separate service or database extension, and behaved consistently in the final test suite and targeted probes.
- What got in the way
- Initial scoring returned 24 results for an exact identifier when the test expected one, so exact-match handling and ranking behavior required adjustment.