Shivendra Tripathi

About Shivendra Tripathi

A highly motivated and adaptable CS undergrad with a passion for Python and ML. I have a…

Experience

1. Senior Data Scientist

   Meta

   May 2025 — Present

   Built large-scale learning-to-rank models for feed personalization using Python and PyTorch, leveraging user behavior signals and embedding- based features, improving CTR and engagement metrics by 12% through optimized feature representation of social interactions. Designed statistically rigorous A/B testing frameworks incorporating hypothesis testing, variance reduction techniques, and causal inference methods to evaluate ranking experiments, improving engagement quality (comments, saves, shares) by 6–9%. Developed transformer-based NLP pipelines (BERT/RoBERTa via Hugging Face) to extract semantic features from multimodal content (text, captions, metadata), improving ranking relevance and reducing low-quality impressions by 15%. Implemented end-to-end ML pipelines on AWS SageMaker and Lambda with automated retraining workflows, feature versioning, and CI/CD integration, reducing model deployment cycle time by 40%. Applied SHAP and LIME explainability techniques to interpret learning-to-rank models for feed personalization, identifying key user behavior and embedding-driven feature contributions, improving transparency, debugging efficiency, and stakeholder trust in ranking decisions. Engineered scalable feature engineering pipelines using Spark for real-time behavioral signals (dwell time, scroll depth, clickstream events), improving prediction accuracy of engagement models. Experimented with reinforcement learning inspired ranking policies to optimize long-term engagement metrics beyond click-based optimization, improving session-level retention by 8%.