This was my first project at the MIT Media Lab. I had two weeks to build something for the fall Members Event — an evening where sponsors walk through the lab and interact with research demos. I wanted to explore how generative models could support collaborative ideation by acting as shared creative substrates, rather than tools that produce finished outputs in isolation.
Meshup allows multiple users to steer a shared latent space in real time. Each participant selects reference images, adjusts influence weights, and watches as the model synthesizes a continuously evolving output that reflects the group's collective direction. Ideas merge and diverge live on screen, so the conversation happens through the model rather than around it.
The demo worked — sponsors gathered around the screen, grabbed references, and started steering the model together. It was a small thing, but watching people who had never met begin riffing off each other's inputs through a generative model felt like a glimpse of something worth pursuing. Meshup investigates how computation can reduce friction in early-stage creative collaboration, letting AI serve as a medium for collective sense-making rather than an answer engine.
