In the summer of 2024, over a beer with my longtime friend Dan Hammer, we started talking about a familiar obsession: how to make Earth data more useful. Dan and I met over a decade ago as Presidential Innovation Fellows in the Obama White House. I was working to open up data for humanitarian response. Dan was building APIs at NASA to make satellite data more accessible. We bonded over a shared belief: geospatial data, in the right hands, could drive massive impact.

That belief has shaped my career. At Ushahidi, I saw the power of geospatial tools in crisis response. Later, I founded Kettle, an insurance company built around the idea that climate change would upend the industry, and that better models, trained on satellite and weather data, could price risk more accurately. We trained convolutional neural networks on that data, spent millions refining them, and built some of the most accurate wildfire forecasts in the industry. 

And then ChatGPT happened.

A New Paradigm

As transformer models began reshaping everything, I asked myself: If this technology had existed when we started Kettle, what would we have done differently? 

Could you apply transformer architectures — the kind powering LLMs — to the Earth observation data I had been working with all along?

Could we make the Earth searchable? 

As it turned out, Dan and Bruno Sánchez-Andrade Nuño were asking the same question. They had just built Clay, one of the most advanced and fully open large Earth observation models (LEOMs) in the world. Organizations were eager to use it, but kept asking for help getting it into production.

That convergence of ideas led to LGND (short for ‘legend’).

Why LGND Exists

LGND’s mission is simple: Make Earth data accessible and actionable for people and AI.

We’re building infrastructure that allows you to query the planet the way you might query an LLM. Think: space + time + semantics.

The potential here is enormous. The World Economic Forum estimates there is $700 billion of untapped value in Earth observation data. Until now, that data has been hard to work with — fragmented, siloed, and locked in formats built for expert analysts, not AI. 

We believe that’s about to change.

Why Now

There’s a transformation underway across the AI landscape. Transformers are turning once-static data, such as text, images, audio, and video, into searchable, composable inputs. And Earth data is next.

Consider this: We’ve collected nearly 200 petabytes of satellite imagery. That vastly exceeds the <1 PB of text used to train large language models, or even the 1–5 PB of imagery behind DALL·E.

Until recently, making sense of that imagery required teams of experts or custom-trained convolutional neural nets. But with the rise of geospatial embeddings, we’re seeing a shift as fundamental as the one from keywords to language embeddings in natural language search.

For 20 years, text search relied on keywords and hyperlinks until it moved to vector embeddings. Geospatial analysis is going through the same shift: from map tiles and pixel classification to geo-embeddings, a new first-order data object that allows AI to reason spatially.

What’s Possible

This shift unlocks a radically more intuitive way to interact with Earth data:

• Ask: “Show me every port in Asia with congestion higher than X over the last month.”
• Detect new construction, crop stress, or deforestation without training a custom model.
• Build applications, agents, or robots that see and understand the world.
• Ask an LLM: “Map all properties in LA with significant fire breaks between them and public parkland.”

No custom model. No weeks of tuning. Just Earth data, accessible through a new interface layer that speaks the everyday language of AI.

What We’re Building

At LGND, we’re building the geospatial AI infrastructure that makes the planet searchable, monitorable, measurable, and understandable — continuously. 

This powers use cases like:

• Travel: Book a hotel on a white-sand beach with no construction in sight.
• Insurance: Underwrite wildfire or flood risk and track changes over time.
• Finance: Track agricultural productivity or renewable energy deployment.
• Logistics: Monitor congestion or verify shipping disruptions.
• Governments & NGOs: Monitor climate, conflict, and infrastructure with spatial awareness.

 And that’s just for starters.

Join Us

Our vision is to make the Earth queryable over space and time, just like any other data source in the AI stack.

If your company, product, or AI application could benefit from tapping into Earth data at scale, or if you want to help shape the future of planetary intelligence, we would love to hear from you.