washington imae

data analyst

Fiverr • ZZP

Jan 2026 - Present • 6 mos

Fraud Detection Pipeline — Payment Analytics Firm, 2026 Client needed 7,000 transactions cleaned for a fraud model. Five columns had gaps. Device types were a mess — same thing entered three different ways by three different people in the same department. The fraud label was the target column, so I locked it down early. Told the client I wouldn't touch it, period. He agreed. The cleaning script ran in two minutes. Imputed the gaps, standardized the device codes, flagged the outliers without deleting them. Trained a fraud classifier on the output. 10% fraud rate, realistically imbalanced, handled without oversampling tricks. One thing I got wrong: I initially flagged all velocity scores above 10 as outliers. Turns out those were mostly the fraudulent transactions. Client caught it on review. I adjusted the threshold, re-ran, delivered same day. No charge for the fix — my mistake, my time. Multi-Sheet Supplier Cleanup — E-Commerce Dropshipper, 2026 Supplier sent a 12-sheet Excel file. Only two sheets had actual product data. The rest were pivot tables, notes, and a pricing chart from 2024. My profiler picked the right sheets automatically — 4,200 products across two tabs, merged into one clean catalog. Biggest headache: prices were in euros on one sheet, dollars on another, and a third had some genius typing "15 bucks" in the price column. Caught it before it hit the client's store. Standardized everything to USD, stripped the text from the price field, matched SKUs where the formats didn't line up. Client went from "I can't upload this" to live in under 24 hours. Aegis Pipeline — Built 2025-2026 Not a client job. This one's mine. I got tired of doing the same cleaning steps by hand, so I built a system that does it for me. Profiles any CSV or Excel file in seconds. Finds structural breaks mid-file. Detects compound fields, abbreviation codes, outliers. Generates the cleaning script, lets me review it, then execute