Leading organizations and efforts on various online taxonomic classification and text extraction problems at the scale of the Amazon catalog and helping build ML infrastructure for training and inference, and automated DL model cost/latency reduction systems along the way.

Leading a team building a high-availability, high ops (>100k) driver identification system in the cloud using Dempster-Shafer/Bayesian machine learning methods and vision-based deep learning methods; the NoSQL infrastructure for Nauto; and a system to merge time series representing GPS data on a large scale in a consistent manner that is not limited by the CAP theorem. The team won the Scylla user award for part of this work.

Led efforts for lossless compression of IMU data achieving 4x reduction over a custom gzipped binary. See https://medium.com/@saboo.rohit for some of this work.

Working on V1 for several components of a stealth mode product in the space of information.

Worked on several projects in Google Robotics and Search -- some of the launched projects are described below.

Consulted for Morphormix (which is now part of Accuray) for segmentation/image registration methodologies/technologies developed by me during my Ph.D.