June 2025 Summaries

GenreX, founded by Malcolm Yang and Yihao Chen, develops AI-powered tools that generate expressive, customizable, royalty-free music for creators such as game developers, filmmakers, educators, and video editors. The company needed infrastructure that could support rapid research, large-scale model training, and reliable deployment without requiring platform changes or extensive manual cloud management. Using Lightning AI Studios and its Bring Your Own Cloud capability with AWS, GenreX ran GPU workloads, tested training variations asynchronously, managed costs, and shared files through a unified system. The team reports validating its product hypothesis in four weeks for less than $10,000, estimating that the process would have taken at least 70% longer otherwise. GenreX aims to broaden access to music creation by reducing the technical, financial, and licensing barriers traditionally associated with producing original soundtracks, including through integrations with tools such as Canva and Adobe Express.

MIT’s Sustainable Design Lab develops AI-driven urban building energy models through its UBEM.io platform to help cities plan retrofits, shape policy, and reduce carbon emissions. Previously, the lab’s sophisticated simulations were slowed by 36-hour training runs, difficult debugging, and onboarding processes that could take up to two weeks. By adopting Lightning AI Studios for on-demand cloud computing and PyTorch Lightning for simplified, scalable multi-GPU training, the team reduced model training time to six hours, cut onboarding time by 90 percent, and shortened prototype-to-production cycles from roughly three weeks to two. The improved workflow enables researchers to focus more on testing and refining city-scale geospatial models, supporting faster development of tools intended to help municipalities decarbonize.

Columbia University’s Paninski Lab developed Lightning Pose, a video-based, semi-supervised deep learning framework that improves animal movement tracking to support neuroscience research on the relationship between behavior and neural activity. Traditional model development would have required months to train and evaluate thousands of configurations, but using Lightning AI Studios for GPU resources, reproducible environments, and large-scale training sweeps allowed the team to launch 1,000 jobs in two weeks and reduce training time by 90%. Built with PyTorch Lightning, the system incorporates labeled frames, unlabeled videos, and motion-based constraints to produce more reliable predictions under conditions such as occlusions and changing camera angles. On mouse pupil-tracking data, it outperformed models such as DeepLabCut in prediction smoothness, pixel error, and correlations between tracked behavior and neural recordings. The platform also enabled easier sharing of complete, reproducible workspaces with collaborators and reviewers, and Lightning Pose is now used by groups including the International Brain Lab to analyze behavior across more than 30,000 experiments.