About the job
At Roblox, we strive to connect a billion people with optimism and civility, and the Safety organization’s mission is to become the leader in civil immersive online communities. We systematically detect, remove, and prevent problematic accounts, content and behavior, and we make Roblox accounts secure and free from compromise. As the Technical Lead for this pod, you will architect and build an industry-leading detection system that operates at a massive scale—processing billions of accounts and identifying recidivism within minutes. You will also be supporting additional use cases for alt detection across the company.
Responsibilities
Lead the technical vision for alternate account detection platform, moving from reactive measures to proactive, near real-time prevention.
Architect high-scale ML systems using Graph Neural Networks (GNNs) and advanced clustering techniques to map relationships across billions of entities.
Solve complex ground truth and training data challenges for adversarial usecases
Build for latency and scale, ensuring that detection happens within minutes of a bad actor’s attempt to rejoin the platform.
Develop innovative adversarial approaches to stay ahead of sophisticated actors who use evolving techniques to mask their identity.
Drive the ML roadmap, identifying opportunities to leverage big data and behavioral signals to improve precision and recall in a high-stakes environment.
Mentor and up-level a pod of high-performing ML and software engineers, fostering a culture of technical excellence and rapid iteration.
Qualifications
Minimum
MS or PhD degree in Computer Science, Machine Learning, or a related field.
10+ years of industry experience in Applied ML, with a significant focus on anti-abuse, fraud, integrity, or identity.
Expertise in Graph Learning: Deep experience with Large-scale GNNs (GraphSAGE, PGB, etc.) and unsupervised/semi-supervised clustering at the scale of billions of nodes.
Proven track record of leading complex technical projects from conception to production-level deployment.
Experience with high-throughput systems: You understand the nuances of deploying ML models in low-latency environments where "time-to-detect" is a critical KPI.
Adversarial mindset: You can think like a bad actor to anticipate how they will circumvent detection and build robust defenses against it.
Preferred
No preferred qualifications listed.