Reddit - Senior Data Scientist, Safety Enforcement

• Own the enforcement lifecycle. Lead a defined abuse or enforcement problem from problem framing and signal discovery through labels and taxonomies, model or rule development, thresholding, rollout, monitoring, and iteration. • Own the enforcement lifecycle. • Discover and characterize abuse. Analyze large-scale behavioral, content, account, and network data to identify adversarial patterns, emerging threats, and gaps in existing mitigations. • Discover and characterize abuse. • Build and evaluate decision systems. Develop models, classifiers, scoring systems, rules, and analytical prototypes using the statistical, machine-learning, or GenAI methods best suited to the problem. • Build and evaluate decision systems • Define decision-quality standards. Create evaluation frameworks that address precision, recall, calibration, coverage, robustness, fairness, user friction, reviewer capacity, and other operational costs—not model performance in isolation. • Define decision-quality standards. • Ship with engineering partners. Partner with Engineering and Safety ML to productionize promising approaches; define data and logging needs, launch guardrails, monitoring and alerting, escalation paths, and rollback criteria. • Ship with engineering partners. • Translate policy into systems. Work with Policy and Safety Operations to convert written policy and product requirements into modelable targets, labels, decision rules, reviewer workflows, and measurable enforcement outcomes. • Translate policy into systems. • Measure real-world impact. Design staged rollouts, experiments, and observational analyses that test whether interventions reduce harm while managing false positives and downstream effects on users, communities, and operations. • Measure real-world impact. • Respond and learn. Support investigations and urgent escalations, identify mitigation options, conduct post-incident analyses, and turn one-time findings into durable detection or monitoring improvements. • Respond and learn. • Scale the team’s effectiveness. Build reusable datasets, evaluation harnesses, monitoring tools, and documentation; communicate clear recommendations to technical and non-technical partners; and mentor junior peers when opportunities arise. • Scale the team’s effectiveness. • Stay ahead of the threat landscape. Track emerging abuse tactics and advances in detection, adversarial ML, language models, synthetic data, and human-in-the-loop safety systems, and apply them pragmatically. • Stay ahead of the threat landscape.