HASH: Building homegrown AI for maternal and reproductive health in Africa

The problem

Poor reproductive health remains one of the largest causes of illness and death worldwide, disproportionately affecting women of reproductive age. In Sub-Saharan Africa, the picture is particularly acute — 1 in 26 women faces a lifetime risk of dying from pregnancy-related causes, and adolescents remain especially vulnerable to sexually transmitted infections, unintended pregnancies, and unsafe abortions

In recent years, health facilities across the region have begun generating more digital data than ever before — patient records, diagnostic results, community health reports. In theory, artificial intelligence (AI) could help put that data to work: supporting health workers to spot complications earlier, flagging which patients are most at risk, or giving people access to reliable health information in their own language. The potential is real and growing.

But for most governments in the region, turning that potential into practice has been difficult. The technical skills needed to build AI health tools locally are scarce. Funding for homegrown research and development is limited. And the vast majority of AI innovation in health has been led from outside the continent — often built for different health systems, different languages, and different populations.

The result is a gap that many working in public health will recognise: the technology exists in principle, but the local capacity to develop, test, and deploy it does not. And without that local capacity, even well-intentioned tools risk being poorly matched to the communities they are meant to help.

The solution

The Hub for Artificial Intelligence in Maternal, Sexual and Reproductive Health (HASH) was established in 2021 to close that gap. HASH acts as an enabler — a programme that finds, funds, and supports African-led teams that are building AI tools for maternal, sexual, and reproductive health (MSRH) in their own countries. Rather than developing a single product or bringing in technology from outside the continent, HASH's role is to help homegrown innovations move from a promising idea to something that can actually be used in a clinic, health facility, or community.

For governments and health systems watching the global AI conversation accelerate, this model offers something distinct. Instead of procuring a finished tool and hoping it fits, HASH supports the development of tools that are built by local teams, grounded in local health priorities, and designed for the languages and contexts where they will actually be used.

In 2022, the hub put out an open call for proposals and received 80 applications from across the region. Of those, 10 were selected from 7 African countries. The projects covered a range of health challenges that public health teams across Sub-Saharan Africa deal with daily and used different types of AI to address them. Four were chatbots — automated messaging tools that help patients navigate self-care and make informed decisions about their health. Four were predictive analytics tools — software that looks at patient data to flag risks early, such as the likelihood of miscarriage, whether someone is likely to take up pre-exposure prophylaxis (PrEP, a medication that prevents HIV infection), or what side effects a patient might experience from a particular contraceptive. Two were image analysis tools — one trained to screen for tuberculosis (TB) from medical images, and another to detect uterine fibroids.

Over 15 months, HASH provided each team with a tailored package of support to help them further develop their tools. This included 16 training sessions focused on building practical technical skills, one-on-one "AI clinics" where teams could bring specific problems and get direct help, three peer learning journeys that brought innovators together across countries to share what was and wasn't working, regular check-ins organised around common themes, and ongoing mentorship from a committee of experts in both AI and reproductive health. For public servants involved in health innovation or digital transformation, the model is worth noting: rather than funding a tool and hoping for the best, HASH wrapped sustained, practical support around each team throughout the development process.

Alongside the innovator programme, the hub invested in building foundational resources that did not previously exist in the region. One was a parallel text dataset called SALT — matched text in five Ugandan languages, designed to help developers build more accurate translation and speech-to-text tools in languages that millions of people speak but that most AI systems do not yet support. Another was an open-access collection of sexually transmitted infection (STI) questions and answers, grounded in African health contexts and gathered through a public crowdsourcing effort, intended to help test and improve the accuracy of AI-powered health assistants being developed across the region.

The impact

1. Ten out of ten projects completed
All ten innovations selected through the open call completed their projects over the 15-month programme. Of those, two reached what the HASH team describes as "ready to scale" — meaning they had been tested enough to be considered for wider use — and both were carried forward into Phase 2 funding. Two more reached the pilot stage, where they were being tested with real users and refined based on feedback; one of these received additional funding from the Bill & Melinda Gates Foundation (BMGF). The remaining six were at the prototype stage, with plans for further development. One of these was also brought into Phase 2.

A 100% completion rate across 10 early-stage teams working in 7 countries is worth noting. It suggests that the support model — combining structured training, direct one-on-one help, peer learning, and expert mentorship — was well suited to the practical challenges these teams were facing.

2. A growing network that has outlived its initial funding
In 2022, HASH established a formal network with its ten original innovator teams as founding members. That network has since grown to over 150 members and expanded its collaborations with other AI for global health networks and organisations. Five member-led clubs have emerged — covering datasets, capacity building, gender equity and inclusion, stakeholder engagement, and scientific communication — each running regular events organised by members themselves.

The network has also run sessions on topics including educating healthcare workers about AI, engaging communities in AI development, and integrating gender, equity, and inclusion into the design and development of AI tools. The fact that this infrastructure has continued beyond the initial funding cycle and now operates under a shared governance model — where decisions are made jointly with the innovator teams, not just by the hub — suggests it has developed enough value and ownership to sustain itself.

3. Language and data resources that did not exist before
Beyond supporting individual projects, HASH invested in shared resources for the region. The SALT dataset provides matched text in five Ugandan languages, giving developers a foundation to build more accurate translation and speech-to-text tools in languages that millions of people speak but that most AI systems do not yet understand. The sexually transmitted infection (STI) question-and-answer collection, gathered through a public call for contributions and freely available to anyone, offers a resource grounded in African health contexts that any MSRH chatbot or digital health assistant in the region could use to test and improve its accuracy.

These outputs address a practical challenge. AI tools depend on the data they learn from. In settings where reliable health information in local languages is scarce, having that foundation in place affects whether tools are accurate and relevant for the communities they are intended to serve.

Lessons learned

• Flexible support matters more than a fixed curriculum. Each innovator team was at a different stage, working in a different country, and facing different challenges. The HASH team found that combining scheduled training with on-demand one-on-one clinics and tailored mentorship was more effective than delivering the same programme to everyone. The ability to adapt support to what each team actually needed was identified as a key factor in all ten projects reaching completion.
• Peer learning across countries proved high value for modest cost. The peer learning journeys and quarterly check-ins brought together innovators from different countries working on different health problems. According to the HASH team, this cross-regional exchange helped teams learn from each other and avoid repeating the same mistakes. Structured opportunities for practitioners to share knowledge directly with each other, rather than relying solely on top-down training, were a consistent feature of the programme.
• Connections to established networks opened doors. HASH's partnerships with AI and global health networks — including The Global Health Network (TGHN) — raised the profile of its innovator teams and connected them to opportunities they would not otherwise have had access to. At least one project secured additional funding from the Bill & Melinda Gates Foundation through these connections. For smaller programmes operating with limited visibility, being part of a wider ecosystem helped ensure that good work did not go unnoticed.
• Local language data shaped the programme from the start. The decision to build resources like the SALT dataset and the STI question-and-answer collection reflects a practical reality: AI tools built using data from other regions, in other languages, do not automatically work well in Sub-Saharan African health settings. Investing in health information that is in the right languages and reflects local conditions was a core part of HASH's approach.

This case study was written with assistance from artificial intelligence.