Science operations scheduling is crucial for any spacecraft to deliver valuable scientific results after launch. Unfortunately, finding the operations schedule that gives the best results is very difficult due to all the constraints present. We introduce a machine-learning framework capable of optimizing science operations scheduling and demonstrate its application to NASA’s upcoming Carruthers Geocorona Observatory mission. During pre-launch stress tests, we successfully demonstrated the ability to schedule three months of operations in under 6 hours, demonstrating real-time responsiveness for mission-critical scheduling.