Fraud and Anomaly Detection ML Intern

Bad actors, small signals

Fake recruiter accounts post listings that ask candidates for a registration fee, and by the time a human reviews the report, forty people have already applied. You will build the model that catches them sooner.

Two months of work

Week one is a guided tour of the abuse data with the trust and safety team. After that you engineer features from posting behaviour, account age, text similarity to known scam listings and payment keyword patterns, then train an anomaly detector alongside a supervised classifier on the labelled reports we already hold. Class imbalance here is severe, so precision recall curves and a threshold you can defend to a non technical reviewer matter far more than an accuracy figure. Your final week goes into a writeup of what the model still misses and why.

What you need

A final year student or recent graduate, comfortable in Python and pandas, with scikit-learn picked up from coursework or personal projects. No professional experience expected. The role runs fully remote within India for two months, with a daily standup and one deep review session each week. If you enjoy reading a dataset until it confesses, this will suit you.