About

Aurko Roy is a Senior Research Scientist at Google, specializing in optimization, machine learning, NLP, and generative models. He has a PhD in Algorithms, Combinatorics & Optimization from Georgia Tech, where he was a student from 2013-2017. Aurko's current work at Google focuses on NLP research and infrastructure, including the development of Lingvo JAX, a library for training large language models, and a 200B parameter model submitted to MLPerf. He has also conducted research on improving Transformer models with latent n-grams and has applied his work to various Google products, such as Ads, YouTube, and Search. Before his current role, Aurko worked as a Research Scientist and Software Engineer at Google, where he focused on applied research in NLP, including machine translation, paraphrase generation, language modeling, question answering, retrieval, and document understanding. He has launched several products, including the Routing Transformer for YouTube and Google Search, which improved recall and F1 scores over Transformer/BERT for video and UGC comments classification models. He also developed the Routing Transformer for Google Docs and Gmail summarization, enabling summarization of long documents and emails. Aurko's education includes a PhD in Algorithms, Combinatorics and Optimization from Georgia Institute of Technology, a Master of Technology in Computer Science from Indian Institute of Technology, Kanpur, and a Bachelor of Technology in Computer Science, also from Indian Institute of Technology, Kanpur. His tech stack includes expertise in Research Scientist, NLP, BERT, Search, ML, language models, LLM, Video Classification, dotNET, Tensorflow, infra, and Software Engineer. With 5.78 years of relevant experience, Aurko is interested in applied research positions in tech and quantitative finance.