Abhishek Majumdar

I am a Staff / Tech Lead Machine Learning Engineer with 9 years of AI/ML experience building large-scale production systems across Search, Recommendations, Ads, Reinforcement Learning, LLM/RAG systems, Dense Retrieval, Ranking, and Graph Neural Networks. I currently work at Meta on Threads Targeting, Bidding, and Relevance, where I drive ML initiatives across reward optimization, causal/offline evaluation, multi-objective ranking, value model simulation, Multi Armed Bandit and long-horizon optimization workflows. My work focuses on translating ambiguous product and business goals into scalable ML systems with measurable impact. Previously at Roku, I led and built ML systems across search, personalization, semantic understanding, query understanding, dense retrieval, knowledge graphs, recommendation systems, and reinforcement learning. My work powered large-scale consumer experiences including voice search, categorical search, personalized retrieval, query suggestions, recommendations, and revenue vs relevance optimizations. My technical interests span frontier AI systems, applied ML research, reinforcement learning, Agentic AI, retrieval and reasoning systems, graph neural networks, personalization and ranking, large-scale model evaluation, and production-scale ML infrastructure. I am an inventor on 14 published patents across reinforcement learning, retrieval optimization, personalized ranking, heterogeneous graph neural networks, LLM transfer learning, LLM-generated training data, and multi-query projection architectures. I also serve as a Core PC Member / Reviewer for top ML Conferences.