Clean and deduplicate a 40,000-row customer database

Project Description

Three systems merged over five years, 40,000 customer records, and we do not know how many are actually distinct people. Same person appears as different rows with a spelling variant, a different phone format, or the same email in different case. Indian names make naive fuzzy matching produce bad merges, so we want a conservative approach with a review queue rather than aggressive automatic merging. Deliverable: the cleaned dataset, the matching rules written down, and the script so we can run it again.

Skills Required