Overworld is building world models - AI systems that understand and generate interactive environments. We're a small team of researchers and engineers, most of us from open-source AI and the games industry, working at the intersection of video generation, real-time inference, and human interaction.
We do our research in the open and publish what we build. We're looking for people who want to do the same.
World models, diffusion, representation learning, RLHF, video generation. Both theoretical work and hands-on implementation.
Pretraining infrastructure, distributed systems, data pipelines. The work that makes large-scale model training possible.
Real-time inference, interactive systems, and the connective tissue between a model and the people using it.
Computer vision, interactive graphics, spatial computing. Bringing world models into physical space.
We're remote-first and distributed. Most coordination happens asynchronously. We value people who can own their work end to end - from figuring out the right problem to solving it to telling others what they found.
Open source is central to what we do. Our research, our data collection tools, and our models are public. You should be comfortable working in the open.
Interested? Get in touch.
Apply