We're a small lab working on how the physical world gets translated into something machines and people can actually use.
While AI systems have advanced quickly, how well they understand any given part of the world still depends heavily on how much data existed about that place to begin with. Some regions and contexts are covered in rich detail. Others are represented poorly, leaving real gaps in what these systems can see and understand.
Earth embeddings are starting to change what's possible here. By compressing enormous, complex Earth data into a smaller, shared representation, they give anyone a starting point that's already global, rather than requiring a model built from scratch. In lower-resource settings especially, this matters: instead of needing large amounts of local data before you can begin, you can start from a representation that already carries global structure, and build from there.
To bridge this gap, OpenEarthStack is doing the work to make Earth embeddings easier to use and adapt for local context.
Working in the open moves this field faster than working in isolation. We plan to publish what we learn as we go, findings, code, and tools, rather than waiting until something is finished. Sharing early keeps us honest, and gives others a head start instead of a repeated one.
Powerful tools mean little if they're hard to use. We care as much about how something feels to work with as we do about what it can technically do, especially for people encountering these tools for the first time.
Data gaps and infrastructure gaps aren't abstract to the people building this, we've lived with them. That's what shapes what we choose to work on, and why local context isn't an afterthought here.