Staff ML Engineer - Trust & Safety

AVIV, Berlin

What you will do

Models across the trust stack: Design, train and ship identity risk scoring, fraud and fraud-ring detection, listing integrity scoring, duplicate and republication detection, and behavioral signals on both sides of the marketplace.
Adversarial resilience and error balance: Build models and retraining loops that hold up as bad actors adapt, catch bad listings before they are reported, and treat false positives as seriously as false negatives, because a false positive is an honest owner who can't publish their home.
Evaluation you can trust: Run careful experiments that account for adversarial drift and feedback loops, where enforcement changes tomorrow's training data, and give an honest analysis when offline and online results disagree.
The human moderation loop: Connect reviewer decisions to model training and route uncertain cases to reviewers, integrating with moderation tooling such as Implio and measuring the whole loop.

AVIV, Berlin

Models across the trust stack: Design, train and ship identity risk scoring, fraud and fraud-ring detection, listing integrity scoring, duplicate and republication detection, and behavioral signals on both sides of the marketplace.
Adversarial resilience and error balance: Build models and retraining loops that hold up as bad actors adapt, catch bad listings before they are reported, and treat false positives as seriously as false negatives, because a false positive is an honest owner who can't publish their home.
Evaluation you can trust: Run careful experiments that account for adversarial drift and feedback loops, where enforcement changes tomorrow's training data, and give an honest analysis when offline and online results disagree.
The human moderation loop: Connect reviewer decisions to model training and route uncertain cases to reviewers, integrating with moderation tooling such as Implio and measuring the whole loop.