Recent developments in GPU hardware and storage technology have changed how we do data analysis and machine learning. These technologies on a single node have grown many folds in the last five years while the growth in network speed has lagged behind. I will talk about the overall ML lifecycle and challenges we face in doing ML at scale, from protecting your Uber accounts to making self driving cars a reality. Then I want to focus on an important part of ML lifecycle which is data/ML exploration and experimentation. In large companies like Uber, data scientists are inclined to use shared Hadoop infra for all their needs. For data exploration, this is inefficient for the user and also makes the cluster run slow. I will talk about our new solution to tackle this problem by using a high powered node that lets us to work with 100s of GB to few TBs of data interactively without paying the overhead of a distributed system. I will also talk about some of the interesting machine learning and infrastructure problems that I face in my new role in Uber’s self driving team.