When people ask us what Medic’s AI strategy is, I give an answer that surprises them: fifteen years of context.
Building for community health teaches you, slowly and sometimes the hard way: the technology was never in shortage. What’s scarce is knowing what a day actually looks like for a health worker, which parts of her day are heavy, what she reaches for when the system fails her. That knowledge doesn’t live in any model, but in a community, and it takes years to earn.
No one holds the whole picture alone
This is the part of our work people outside it are most surprised by. Medic stewards the Community Health Toolkit (CHT). But we didn’t build Zanzibar’s national community health program; D-tree did, together with the government, over more than a decade. They took the CHT and made it the backbone of Jamii ni Afya: a program that grew from three modules to fourteen, from a volunteer effort to a paid government cadre. And they shaped the platform as they went. The CHT that exists today carries D-tree’s field lessons inside it, the way it carries lessons from our partners in Kenya, Nepal, Mali, and everywhere else it runs around the world.
That’s what a community of practice means, and it’s what the second Ground Truth conversation, with Abbas Wandella of D-tree, kept bringing home for me. Medic stewards the platform, which means keeping it open, but also building the community around it and turning one country’s lessons into everyone’s progress. Implementers like D-tree hold the ground: the government relationship, the trust of health workers, the knowledge of what the last day of the month actually looks like for a supervisor in Unguja. The tool gets better because the people using it can change it, and those changes flow back to everyone. When Abbas explains what made national scale possible, he doesn’t start with features, but with open standards: that’s what gives you adaptability, and what makes it possible for a government to scale a solution and truly own it.
Meeting people where they are
Our conversation kept circling one discipline: understand the problem before you reach for the technology. Since our founding days, Medic’s core philosophy has been “We don’t start with technology; we start with people.” Human-centered design is how that philosophy became a practice. The CHT was built sitting with health workers in Togo, Nepal, and Kenya, designing with them, and it’s still how we work today, including the community squads formed around the real needs of the people using the tools. D-tree’s work in Zanzibar reflects that same ethos, a decade deep.
When D-tree’s team researched what supervisors actually needed, they found that supervisors weren’t short on data; they were drowning in it. They spent hours every month compiling and reconciling data, and the challenge was getting useful insights before they could start their actual job: supporting their team. What they built follows from that, and every choice they made meets people where they already are. The assistant lives on WhatsApp, where supervisors already work. It speaks Kiswahili. It answers from the Jamii ni Afya data supervisors and their teams already collect and trust. It draws on ministry-approved protocols and is designed to guide, not decide, so the judgment stays with the human.
The discipline goes deeper than design choices. D-tree’s research surfaced seven possible uses for AI, and they moved forward with only one. The problem they chose was, in Abbas’ framing, worth solving with or without AI. This sentence really stayed with me since the conversation. If a problem isn’t worth solving without the model, it isn’t worth solving with one.
The road before the last mile
We feel this on the platform side too. Before any model can serve this community well, someone has to have captured how care actually happens: the workflows, the languages, the constraints, the relationships between a health worker and the families she serves. Fifteen years of that knowledge lives in the CHT community, contributed by everyone at the table. That isn’t the preamble to the AI work. It is the work. The model is the last mile of a road the community spent years building together.
Responsible AI for community health isn’t a technology. It’s a community that put in the years before the model. Research before building. One use case instead of seven. The person’s language, on the tools she already holds. The model arriving last, for a problem worth solving anyway.
Trust is built years before any tool arrives. Our job, all of us in this community of practice, is to keep building it together, so that when the right tool comes, it has ground to stand on.