San Francisco Bay Area
Meta Recommender Systems -> Core Modeling -> Retrieval ●Driving the transition from traditional pipelines to LLM-native recommender systems across multiple Meta products (e.g., Threads, Instagram), including pretraining, semantic ID generation, token expansion, and post-training. ●Designing and deploying large-scale neural retrieval systems, introducing learnable indexing architectures (neural codebooks) that replace ANN-based pipelines and improve retrieval quality and scalability. ●Developing cascaded decoder-based LLM architectures for recommendation, combining retrieval and generation to enable personalized item selection and generative discovery.
Superintelligence Labs -> Search Team ● Contributed to the development of Meta AI's first external search system, focusing on LLM post-training, alignment, and evaluation for high-quality user-facing responses. ● Designed and deployed LLM-as-a-Judge frameworks (Factuality, Helpfulness) and applied DPO-based post-training to improve alignment with human preferences, including multilingual optimization. ● Led human–LLM alignment studies to calibrate automated evaluation systems against human judgments, establishing reliable evaluation pipelines for model iteration. ● Built core modeling components for response quality, including a freshness classifier and public-figure recognition model integrated into production safety and compliance systems. ● Owned measurement and evaluation strategy for multilingual launches, and applied distillation techniques to scale LLM-based evaluation with reduced latency.
Monetization -> Ads -> GenAI ● Led end-to-end development of the \textbf{Financial \& Product Services Classifier} for CFPB regulatory compliance, spanning feature engineering, model training, infra design, enforcement-stack and frontend integration, and revenue impact monitoring; now deployed in production across all U.S. Ads.
Fraud Detection Team ● Specialized in Representation Learning with a keen focus on Gestural Data, driving innovative ways to interpret human movements. ● Conducted comprehensive exploration of Contrastive Learning based on supervised and self-supervised approaches for Anomaly Detection. ● Engineered a Multi-Modal Architecture tailored for Gestural Data, seamlessly integrating images (Computer Vision) and text data (Natural Language Processing). ● Successfully incorporated Contrastive Learning into a Multi-Modal setting, enhancing the overall effectiveness of data analysis and interpretation. ● Demonstrated expertise in implementing diverse ML architectures such as ResNet, AlexNet, and Convolutional AutoEncoders.
Advertisement Fairness Team ● Conducted extensive research on mitigation techniques to ensure fairness in model outcomes with respect to gender and race, contributing to a more equitable AI approach. ● Innovatively implemented a new loss function to effectively improve miscalibration concerning protected attributes, enhancing model fairness. ● Accomplished significant improvements in subgroup calibration with respect to gender within production-level Machine Learning models, ensuring more balanced performance. ● Developed and successfully integrated code in multiple languages and frameworks including C, Python, Caffe2, and PyTorch, demonstrating wide-ranging technical expertise.
Privacy Team ● Contributed to a series of research collaborations with the esteemed team at CSIRO, focusing on cutting-edge issues of Location Privacy. ● Authored multiple papers recognized and published in prestigious forums including the IEEE International Conference on Computer Communications (INFOCOM) and IEEE Transactions on Knowledge and Data Engineering (TKDE). ● Actively collaborated with industry partners, playing a crucial role in integrating privacy-preserving algorithms to bolster data sharing practices across their organization.
● Applied expertise in Supervised Machine Learning methodologies, implementing complex models like Logistic Regression, Random Forest, and Decision Tree to drive insights. ● Utilized Unsupervised Machine Learning algorithms, including K-means, to extract insightful patterns and relationships from data. ● Demonstrated proficiency in organizing Relational Databases, leveraging SQL to manage and manipulate complex datasets.