Using AI to map hope for refugees with UNHCR, the UN Refugee Agency

Refugee camps are complex, dynamic ecosystems that often are stood up quickly and develop organically to meet pressing needs. As these settled areas grow, humanitarian aid organizations such as UNHCR, the UN Refugee Agency, struggle to deliver even the most basic infrastructure and services. In the case of protracted crises when camps become long-term shelters — like the Kalobeyei settlement in Kenya, which shelters more than 300,000 refugees from over 20 countries — the challenge of creating something that functions like a city can feel insurmountable.

At the core of any populated area is spatial data: maps, grids, and reliable location information about schools, hospitals, homes, and roads. Yet in many refugee camps, such data doesn’t exist. There are no formal street names, no consistent addresses, and no clear way to track where essential resources are located. Without this foundation, urban planning becomes guesswork.

So where do you begin?

This was the problem confronting the UNHCR, and other humanitarian aid organizations supporting and empowering people forced to flee their homes. Their hypothesis was bold: perhaps artificial intelligence and open source technologies could provide not only the technical tools, but also the collaborative framework and global community needed to tackle the seemingly impossible.

That’s where GitHub came in: the home for open collaboration and the hub that could connect local knowledge to a global network of problem solvers.

From drones to data

The work began with the people who know Kakuma best: the refugees and residents themselves. With training from the Humanitarian OpenStreetMap Team (HOT), they flew drones over the settlement, capturing thousands of images. Then, they annotated a portion of those images by hand — marking homes, solar panels, clinics, sanitation facilities, and more.

With this “ground truth” in place, the Microsoft AI for Good Lab trained machine learning models to recognize similar features across the rest of the camp. What would have taken years to map manually could now be charted in a fraction of the time.

Turning code into collaboration

But the breakthrough wasn’t just in how the data was gathered. It was in how it was shared. 

Every dataset, model, and line of code was published openly on GitHub. That meant the tools built for Kakuma could now be adapted for other refugee camps, disaster recovery zones, or rapidly growing cities.

GitHub Copilot helped streamline the process, saving hours of formatting and cleanup so the final repositories were clear and ready for anyone to use.

For a deeper look at how UNHCR, The Hive, Microsoft, and partners made this possible, see the Kakuma and Kalobeyei: Drone Imagery and Machine Learning for Better Planning of Refugee Settlements project >

Why GitHub matters

GitHub wasn’t just the publishing platform. It was the hub that connected developers, data scientists, and civic technologists to UNHCR’s mission. With open source code available, others could build on the project, improve it, or apply it to new contexts.

As one project lead explained:

“GitHub is the beginning for the next phase of this project. Connecting developers to these kinds of problems is our mission.”

Building hope together

At its heart, this project is about collaboration. Refugees provided the data, with training and support from the HOT. Scientists and developers ensured the tools were reusable and shareable. Together, they created a blueprint that can guide sustainable planning in Kakuma — and in communities around the world.

For UNHCR, the project was about creating knowledge and tools to support millions of displaced people with dignity and foresight.

And thanks to GitHub, that knowledge doesn’t stay locked in one place. It lives in the open, where anyone, anywhere, can use it to build not just maps but hope.