SolarSprint

Headshots
The Problem
Our Solution

About the project

California throws away a lot of solar power. On sunny days the state produces more than its power lines can carry, and the extra is curtailed. That rose 23% from 2023 to 2024, and about 70% of it is caused by congested lines, not low demand.

Our advisor, Prof. Benjamin Lee, showed that putting AI data centers right next to solar farms can put that power to use. The hard part is that free solar comes and goes. Our project is the software that decides which computing jobs run when, using solar, a battery and a limited grid connection, while still meeting every deadline. Flexible jobs like model training move into the sunny midday window, and must-run jobs like a chatbot get covered by the battery at night. If a cloudy forecast makes a deadline impossible, the system says so up front.

We started the semester working on household energy scheduling but found it was already well covered by research and existing companies. After meeting with Prof. Lee, we switched to this problem. It's newer and bigger, and it ties into his recent work.