For five years, a single idea dominated Chinmay Shrivastava's thoughts: humans didn't have enough hours in the day to build the designs of the future. An avid lover of science fiction, Chinmay struggled to see how current mechanical engineering methods could build the starships that he dreamed would take him throughout the galaxy. While studying computational mechanics during his undergrad at IIT Roorkee, he watched a TED Talk in which Maurice Conti used AI to optimize a car chassis through millions of iterations. The resulting design outperformed what any human could have created alone.
But real-world engineering doesn't work that way, yet. High-fidelity simulations take days, cost thousands, and require hours of expert setup. Engineers are limited in how often they can test and improve their designs. Immediately after watching that TED talk 5 years ago, Shrivastava began working on AI models to accelerate simulation, but the early approaches, CNNs and LSTMs, couldn't properly model nonlinear physics or generalize beyond their training data.
