From Sequential to Parallel: Story about Bayesian Optimization

In 2017, Riskified created its first training platform, the goal was to manage the full model life cycle -from the data set creation to model evaluation. My role in this project was to write the training component. The platform was a great success and helped the research team and Riskified grow. However, the Riskified of 2017 was not the Riskified of 2019 and what we trained in one month in 2017, in 2019, we can train in one day. In 2019, Riskified decided to take its training platform to the next level and created a multidisciplinary team to make it happen. The team included 2 software engineers, a data engineer, and myself as the data scientist.