Shanghai, China
Led one of PayPal's most critical risk domains, identifying fraudulent sellers at scale. Delivered first-ever near-real-time event-sequence embedding for real-time modelling and first-ever LLM-based seller business model detection. Pioneered LLM integration in risk models since 2023–2024. Built PayPal's first two internal LLM use cases for productivity tooling and legal document research. Currently developing an agent framework for dispute resolution, appeal handling, and compliance workflows (from Aug 2025).
Built scam prevention intelligence from zero to one. Refined scam definitions from vague business requirements, aligned objectives across multiple stakeholders, and designed the full solution framework: tagging enhancement, real-time detection model, and near-real-time account-level detection. Led a team of 6 to define and formalise the solution framework for PayPal's resolution ML suite — including instant buyer payout, abusive buyer detection, chargeback protection for merchants, and seller protection payout control.
Developed core identity-related intelligence for fraud prevention — including account quality assessment, account cohesiveness scoring, network-based detection of fraudulent onboarding, and new data onboarding for fraud use cases.
1. In chrage of the weekly tracking of performance of the eBucks; 2. In charge of the measurement of the Seller Trigger Email campaigns’ performance (iGMV, iLSTR…) and accomplishing the automation of the SQL code; 3. In charge of the measurement of performance of the social media ads daily and weekly including importing data via Facebook API with Python and publishing dashboard with tableau.