Probabilistic record-linkage platform from my 4th-year internship at Office des Changes: a 12-step Splink 4 pipeline that deduplicates banking-transaction records, with a Flask + React human-in-the-loop review plane. Blocking cut ~200M candidate pairs to ~90k and resolved 97% of ambiguous matches on the first pass.
react docker flask internship record-linkage entity-resolution data-deduplication human-in-the-loop 4th-year duckdb splink fellegi-sunter office-des-changes
-
Updated
Oct 19, 2026