Data Engineering · Data engineering
Address Intelligence
Large-scale fuzzy entity resolution pipeline
Built an address-matching pipeline for records that could not be joined with exact text matches. It cleans address fields, narrows possible matches by ZIP or geography, then uses RapidFuzz similarity scores to find likely matches at multi-million-record scale.
Processed more than 5 million address records.
Normalized punctuation, abbreviations, and inconsistent text before matching.
Grouped records before fuzzy matching so the system did not compare every address against every other address.
Used similarity scores to rank likely matches instead of requiring exact string equality.
Passed the matched data into downstream analytics workflows.
Comparing every record with every other record would be far too expensive at this scale, so narrowing the candidate set was essential.
Small formatting differences can describe the same address, which made data cleaning directly affect match quality.
Turned millions of inconsistent address records into a dataset that could be matched and analyzed reliably.