Designed and owned a two-stage retrieval + ranking system for personalized account recommendations, engineering behavioral embeddings over distributed graphs to improve successful recovery by 13% (A/B validated).
Built a Mixture-of-Experts model for real-time entity-level prediction at account creation, routing behavioral signals to specialized expert subnetworks and replacing a 14-day batch pipeline with low-latency online scoring.
Led the platform-wide ML roadmap for account re-engagement on Facebook, consolidating 3 legacy pipelines into a unified multi-task ranking architecture (+20K DAU, +1.5M MAU).
Served as AI lead across the org — shipped on-call triage and root-cause agents, and an experimentation agent automating A/B test setup, monitoring, and reporting (saving ~10 hrs/week/eng).
Designed and owned a two-stage retrieval + ranking system for personalized account recommendations, engineering behavioral embeddings over distributed graphs to improve successful recovery by 13% (A/B validated).
Built a Mixture-of-Experts model for real-time entity-level prediction at account creation, routing behavioral signals to specialized expert subnetworks and replacing a 14-day batch pipeline with low-latency online scoring.
Led the platform-wide ML roadmap for account re-engagement on Facebook, consolidating 3 legacy pipelines into a unified multi-task ranking architecture (+20K DAU, +1.5M MAU).
Served as AI lead across the org — shipped on-call triage and root-cause agents, and an experimentation agent automating A/B test setup, monitoring, and reporting (saving ~10 hrs/week/eng).