Roland F. Ganafa-AI Policy and Practice In Uganda

1. What is your name?

My name is Roland F. Ganafa. I'm the Co-Founder and CEO of AI Studio Uganda, and I also serve as Developer Community Lead at Africa's Talking Uganda.

2. Where did you go to school?

I studied at Makerere University Business School here in Kampala. But honestly, a lot of my real education happened outside the classroom, in developer communities, in government offices, and in the field deploying technology where it's actually needed.

3. What did you study?

Business Computing. It turned out to be the right mix for what I do now, enough technology to build, enough business to understand why you're building and who pays for it. Most of the AI-specific knowledge came later, through self-study, community, and doing the work.

4. From your LinkedIn we learned that you have been working in AI. Tell us more about AI Studio Uganda?

AI Studio Uganda is an applied AI innovation studio. The key word is applied. We don't do AI for the sake of demos, we build systems that solve Ugandan problems and we deploy them where Ugandans actually are.

We work closely with the Science, Technology and Innovation Secretariat under the Office of the President, which anchors our work in Uganda's national innovation agenda. A few examples of what we build:

• Ease Health - an offline AI clinical decision support tool for health workers, built on a fine-tuned medical language model. We serve as implementing partners of Crane AI Labs on this, running field research in Luweero District with co-investigators from Makerere University Hospital. It works without the internet, because that's the reality in most health facilities.

• AskCrane - a civic AI assistant that runs on WhatsApp and USSD, designed to connect citizens to services across seven government agencies through NITA-UG's infrastructure. USSD matters because the majority of Ugandans are on feature phones.

• AI Studio Academy - our training arm, where we run training sessions with Adaption Labs, university tours, and weekly Build Nights with Africa's Talking. Our scientist serves as an Adaption Labs Ambassador, which gives us compute support to award to developers every month over six months, and we channel that directly through Build Nights to grow the next generation of Ugandan AI builders.

5. We understand that you are really focused on local solutions. Why is this so important? How does this merge with your work with enterprises, institutions and the public sector?

Because imported solutions keep failing here, and we keep acting surprised. A clinical decision tool that assumes constant connectivity is useless in a rural health centre. A chatbot that only speaks English excludes most of the country. A fintech product designed for credit cards doesn't fit a mobile money economy. Context isn't a nice-to-have, it's the difference between a tool that gets used and one that gathers dust.

And here's the thing: local doesn't mean small. It actually makes the enterprise and public sector work stronger. When we work with a company like when we work with enterprise clients in the energy sector, or with the government through NITA-UG, what we bring is exactly that contextual depth, we understand the infrastructure constraints, the languages, the regulatory environment, and the way Ugandans actually access services. Enterprises and institutions don't need another generic AI vendor. They need a partner who knows that the last mile in Uganda runs on USSD and mobile money, not broadband and credit cards.

6. You stated that "Africa cannot afford to be a passive consumer of AI built elsewhere. We need to own the infrastructure, the data, the models, and the institutions that shape how this technology lands here." Help us make sense of this.

Look at the current pattern. African data, our languages, our images, our behaviour, gets collected, shipped abroad, used to train models we had no say in, and then those models get sold back to us as products. We provide the raw material and buy back the finished goods. That's an extractive economic structure, and we've seen this movie before with other resources.

So when I say we need to own those four things, I mean it practically:

Infrastructure - where does the computation physically happen? If every AI query from Uganda is processed on servers abroad, we have no leverage, no data protection guarantees, and we export value with every request. That's why we host our deployment work on African cloud infrastructure and why I'm interested in sovereign computation.

Data - whoever holds the data decides what the models learn. If Ugandan data only exists inside foreign companies' training sets, our realities get represented on someone else's terms, or not at all.

Models - we need the capability to fine-tune, evaluate, and eventually build models ourselves, not just consume APIs. It's the kind of work our partners at Crane AI Labs are leading from here in Uganda.

Institutions - policy, standards, procurement rules, research bodies. Technology lands the way institutions allow it to land. If we're absent from those rooms, decisions about AI in Africa get made without Africans. It's why I engage with the UN Global Dialogue on AI Governance and why our work with the STI Secretariat matters as much as any product we ship.

None of this means isolation. We partner with Crane AI Labs, IBM, NVIDIA. The point is to partner from a position of capability, not dependence.

7. What is the state of AI adoption in Uganda? Where do we stand on models, institutions, infrastructure and data?

Honest answer: early, uneven, but moving faster than people outside the country realise.

Models - almost nothing is trained here yet, but Ugandans have proven we can shape global models. Our partners at Crane AI Labs built the Uganda Cultural Content Benchmark, which has been adopted by Google DeepMind, OpenAI, and Anthropic to evaluate how well their models handle Ugandan content. Crane AI Labs is featured on Google DeepMind's Gemmaverse page, and they serve as a Gemma Trusted Tester. Their work has also shown fine-tuned open models can run offline for real use cases like clinical support. So the capability exists in Uganda, it needs scale and investment.

Institutions - this is actually a bright spot. The STI Secretariat under the Office of the President and Ministry of ICT and National Guidance are taking innovation seriously, NITA-UG has built genuine digital public infrastructure with UGHub and UGPass, and the EAC Regional AI Strategy now names Project Crane as a flagship sovereign AI intervention for Uganda. The policy conversation here is more alive than people give it credit for. The gap is between strategy documents and budgets.

Infrastructure - our biggest constraint. Compute is scarce and expensive, connectivity is uneven, and power reliability still shapes what's possible. This is exactly why we design offline-first and channel everything through USSD and WhatsApp. You build for the infrastructure you have while pushing for the infrastructure you need.

Data - Uganda's languages and cultural knowledge are barely represented in any major model. We're attacking this directly: our Multicultural Benchmark of Riddles project is recruiting contributors across more than a dozen Ugandan language communities, Acholi, Ateso, Lugbara, Lusoga, Runyankole and others, to create evaluation data that simply doesn't exist anywhere else. Data is where ordinary Ugandans can participate in AI most directly.

On adoption itself: students and developers are adopting fast, we've seen it across the NSSF Career Expo tour through fourteen universities this year. Enterprises are curious but cautious. The government is engaging earlier than most expected.

8. You have done quite a lot of work with the government. Why are business/government interactions so valuable? How do these relationships differ from your technical partnerships?

In a country like Uganda, the government is the largest service provider, the largest data holder, and the rule-setter. If your mission is AI that reaches ordinary citizens, health workers, farmers, students, there is no path that doesn't run through the government. A citizen-facing AI service is only as useful as its connection to the agencies that actually deliver services, which is why AskCrane is built on NITA-UG's integration layer rather than around it.

The two kinds of relationships do completely different jobs. Technical partnerships, Crane AI Labs, IBM, NVIDIA, give us capability: model access, compute, training resources, global credibility. Government relationships give us legitimacy, distribution, and alignment with national priorities. A partnership with DeepMind doesn't get an AI tool into a district health centre. An MOU with the Ministry of Health, Makerere University Hospital, IDI can.

They also run on different clocks and different currencies. Technical partnerships move fast and are transactional in a healthy way, you demonstrate capability, you get access. Government relationships are slower and built on trust, consistency, and showing up repeatedly. You can't parachute in with a pilot and disappear. But when they work, they're the difference between a demo and national infrastructure.

9. Tell us about your work with language models?

It runs on a few tracks.

First, evaluation, making models accountable to Ugandan reality. We work closely with our partners at Crane AI Labs, who built the Uganda Cultural Content Benchmark. It tests whether frontier models actually understand Ugandan context; it's now used by Google DeepMind, OpenAI, and Anthropic, and it was selected for Kaggle's inaugural Task Tuesday. The follow-up, the Multicultural Benchmark of Riddles, goes deeper into indigenous knowledge: riddles are dense with cultural reasoning, and contributors are collecting them with native speakers across Uganda's regional language communities. If a model can handle an Acholi or Lusoga riddle, it understands something real about how we think.

Second, fine-tuning open models for deployment. Ease Health is built on a fine-tuned 4-billion-parameter medical model small enough to run offline on modest hardware. That's a deliberate philosophy: in our context, a small model that works without the internet beats a giant model that needs a data centre.

Third, applied systems - AskCrane and Crane AgriAdvisor put language models behind WhatsApp and USSD so the technology meets people on the phones they already own.

And underneath all of it sits the low-resource language problem. Ugandan languages are nearly invisible to these systems today. Our benchmarking and data work exists to change that, so the next generation of models speaks to Ugandans, not just about them.

10. What are your thoughts on the technology and innovation ecosystem in Uganda? What role can AI play in the bigger picture?

The ecosystem has more talent than capital, and more energy than structure. I see it every week, at Build Night Uganda, at the Deep Tech Summit, across the university tour. The builders are here. What's thin is patient funding, procurement pathways that let startups sell to the government, and the bridge between a brilliant prototype and a sustainable company. Too many of our best builders end up doing outsourced work for foreign clients because the local market hasn't been opened to them.

On AI's role: I'd frame it as a leapfrog opportunity with a deadline. Uganda skipped landlines and went straight to mobile; skipped bank branches and went to mobile money. AI offers a similar jump, one health worker supported by a clinical AI tool extends scarce medical expertise; one farmer with an advisory service on USSD gets agronomic knowledge that never reached the village before; government services become navigable in your own language. AI lets a country short on specialists multiply the specialists it has.

But the deadline part matters. The window in which Africa can shape this technology, contribute data on its own terms, build local capability, set policy, is open now and it won't stay open. If we wait, AI becomes one more thing that happens to Uganda. If we move, it becomes something Uganda builds. That's the whole reason AI Studio Uganda exists.