GPay India serves millions of users daily , which makes us a target for scaled fraud attacks. Proactively preventing frauds at scale with high accuracy becomes critical to maintain safety as well as smooth user experience. This is a particularly complex problem to solve, especially due to the adversarial nature of fraud patterns and very limited labeled data unlike other machine learning problems. One of the ways we solve it in Gpay India, is through a graph based machine learning platform, to prevent repeat offenders from onboarding/transacting on the system. The presentation will discuss about : (1) Designing fraud mitigation enforcements using a large scale graph while also solving for tail end false positives to avoid any good users getting impacted (2) Case study on proactive loss mitigation from fraud ring attacks (3) Designing a feedback loop for continuous model improvement (4) User journey of proactive warnings/nudges to the users while transacting basis graph insights