Led social graph representation learning using GNNs and Transformer encoders, significantly improving friending quality and large-scale retrieval performance.

Explored generative retrieval via semantic tokenization (Semantic IDs) and next-token recommendation models.

Built production LLM pipelines for safe user recommendation, curated high-quality human-labeled datasets, and optimized precision–recall tradeoffs. Investigated preference optimization (GRPO-style) to align open-weight LLMs with policy objectives.

Rebuilt distributed graph training infrastructure (3× cheaper training, 20× cheaper inference), contributing to large-scale open-source GNN systems.

Led development of Pinterest’s first neural retrieval model for Shopping Ads, evolving the system from heuristic interest matching to GraphSAGE-based user representations and a two-tower deep retrieval architecture.

Migrated major recommendation surfaces to NN-based retrieval, driving substantial improvements in engagement quality and monetization efficiency.

Conducted research on CNN-based similar texture image retrieval for Adobe Stock, focusing on developing innovative solutions to enhance the precision and efficiency of image similarity searches.

Conducted research on leveraging deep reinforcement learning for autonomous driving, benchmarked several Deep RL algorithms on ViZDoom