I'm a machine learning scientist in Mountain View, California. For the past nine years I've built ML for education — first at Gradescope as one of two ML scientists behind its grading automation, and now at Turnitin, where I tech-lead AI writing detection and built several of the LLM features in Clarity, a TIME Best Invention of 2025.
Most of my work sits where being wrong is expensive. A recommender can miss a third of the time and still be useful; a system that flags a student's writing, or matches a handwritten name to a class roster, cannot. High precision, false-positive rates that stay low and even across groups, models that know when to abstain — that's the thread through most of what I've built, and it's why I keep coming back to integrity problems. Accountability and safety, a small step at a time.
I like staying close enough to the product to hear from the people using it. This site is where I collect public work and the occasional write-up. Say hello at srbhchandra@gmail.com.
Data
Manchester United's lost decade, in expected goals
The chances United created and conceded since 2014, against goals, points, managers and the rest of the Big Six. Interactive.
Apps
Football Tactics Board
An interactive tactics board: pick a formation, drag positions and players around the pitch, load any Premier League squad, and export the board as an image.
Travel
Japan in 12 days
Tokyo, Nikko, Mount Fuji, Kyoto, Nara, Osaka, Hiroshima and Miyajima by train, and the people who made it the best trip of our lives.
Central Europe in 14 days
A road trip from Prague to Venice through seven countries, day by day: the route on a map, photos, the highlights, and what I'd skip.