Meta album artwork

Meta

ML Engineer

Ads Integrity

Sep 2022 – Oct 2023 · 5 tracks

Menlo Park, CA

Detecting rare, high-severity harms.

Machine learning work on ads integrity, connecting model development with annotation workflows, policy decisions, and real-world detection coverage.

Selected work

  1. From software engineering to ML

    Transitioned from Software Engineering to ML Engineering, specializing in high-severity, low-prevalence ad detection for DMAT (Don't Miss a Thing) system

  2. Stronger enforcement of critical harms

    Improved model action rate by 15% and doubled enforcement for critical harm categories including human exploitation, housing discrimination, and racial discrimination

  3. Better datasets and annotation workflows

    Led cross-functional collaboration with human labelers and product managers to optimize datasets and model performance, directly managing annotation workflows and UI/UX improvements

  4. Correcting political ad detection metrics

    Identified and resolved critical bug in Political Ads detection system that was inflating accuracy metrics; implemented corrected detection thresholds that maintained performance while improving real-world detection coverage in time for 2020 Election.

  5. Policy and model integration

    Partnered with executive stakeholders on policy development for nuanced harm detection scenarios and feature engineered embeddings for multi-model integration