Committed to Innovation: The Master of Science in Data Science Program at the University of San Francisco

Here at Metromile, we are all about making car insurance more fair and less painful — some may call this a lofty goal. We know that all it takes to make auto insurance simple and seamless is a bit of data, some science, a lot of technology, and a dash of magic. From the outside looking in, revolutionizing the auto insurance industry may seem easy, but take a peek under our insurance hood, and you’ll quickly realize that it takes a lot of systemization to keep this engine running. And at a company whose whole mission is to reinvent ways to manage risk, it should come as no surprise that our Data Science team is constantly buzzing; building and testing new models and furiously working to analyze Metromile data to find new opportunities to fix how insurance is traditionally “done.”

pay-per-mile car insurance

Revolutionizing an antiquated industry is no easy feat. It takes a village – a village of brilliant humans that are constantly iterating and innovating. One of the best ways to foster a culture of innovation is to work with local University programs like The Master of Science in Data Science (MSDS) Program at the University of San Francisco. It’s a win-win for both us, and the University because it gives students valuable business experience, and it gives us insight into new data patterns, trends, and opportunities.

The Program

The Practicum Program at USF pairs students with bay area companies, allowing them to apply their skills to gain experience, and reconcile mathematical theory with business practice. Each student is expected to create and refine a project with their partner company for 16 hours per week, while concurrently taking classes. This past Winter and Spring, we were lucky enough to have two USF students join us, Chenxi and Fang. They have spent the past six months trying to read mileage from odometer pictures, which can be used to correct the mileage measurements we receive from the Pulse device. This gets to the crux of what makes Metromile different from traditional insurers – we charge for insurance by the mile, so exact mileage is very important, and we are always looking for new ways to track mileage! Chenxi and Fang utilized some deep learning techniques, like the U-net model, in order to deal with object segmentation problems in computer vision.

The Project

During their time at Metromile, Chenxi and Fang applied state of the art techniques to real-world problems and gained experience using deep learning to solve computer vision challenges. Asking Chenxi and Fang what the most challenging part of their project was, they posited that the most taxing aspect was also what they learned the most from, “trying to solve a problem with limited resources and a relatively small dataset. We tried various ways to enlarge the dataset we have, as well as adjust the algorithms we used to overcome the issue.”

When asked whether or not they experienced a breakthrough moment, both Chenxi and Fang concluded. “There were several breakthrough moments during our work on this project, but the process is gradual and requires constant inputs and modification. The biggest challenge of our practicum was the project itself – extracting mileage from the odometer and correctly distinguishing that information from other similar numbers has turned out to be difficult. Luckily, our mentors have been patient and resourceful and have helped us a lot.”

One of these mentors, Chetan Ramaiah – a Data Science Manager here at Metromile – oversaw the internship program and recalled that Metromile had previously hired a 2016 participant of the program. It was the success from previous years that encouraged him to place Metromile in the program again. “The experiments conducted by Chenxi and Fang helped us identify a state of the art solution to a difficult research problem, and the process helped identify the various difficulties in modeling a solution to the problem. In addition, both Chenxi and Fang helped establish a benchmark on the quality of internship candidates.”

Internships at Metromile

According to Chetan, the project was a success.“The project adds a new dimension to Metromile’s mileage tracking abilities. The odometer project can independently verify and improve our mileage tracking from the Pulse and the OBD-II port.” Metromile looks forward to participating in the Practicum in coming years, and plans on taking on more Data Science and Engineer interns next summer. If you are interested in applying for an internship with Metromile please contact David Clifford, Director of Data Science ( or Mike Dicarlo, VP of Engineering (