The Graduate Who Brought AI to Sierra Leone's Ministry of Education

Mohamed Alpha applied for his Master’s scholarship four times before he got in. Three rejections, then a place at Carnegie Mellon’s campus in Kigali, funded by the Mastercard Foundation, where he trained to be a researcher and engineer in machine learning, wireless communications, energy modelling, and robotics. When the chance came to put that training to work for the Ministry of Education in Sierra Leone, the country he is from, he took it. “It’s a way to give back to Sierra Leone,” he says.

His brief was to coach the ministry’s team and build them an AI tool. Together, the team discussed the problems they faced, and ultimately the team asked for a chatbot, running on WhatsApp, that staff could ask for information e.g. a figure, the number of teachers, the number of pupils, an examination result. Mohamed built it, it works, and the team uses it.

The lasting lesson Mohamed took from the work is about everything underneath the model. The ministry had no central store of clean, current data, often the hardest thing to get hold of, and patchy computers and connectivity to run anything on. His firm view, held from the first week, is that the foundations have to come first.

Before an AI tool,” he says, “you need the infrastructure. Without those things, the AI tool is solving the wrong problem.”

Over six months and across six countries, six ministries of education joined EdTech Hub’s Ministry of Education AI Challenge. Each Ministry worked with embedded AI experts to build their first AI prototypes. Voices from the Ministry of Education AI Challenge introduces the people behind those tools: who they are, how they think, and what it takes to build AI a ministry of education can actually use. For the third profile, we’re featuring Mohamed Alpha in Sierra Leone.

The conversation below is drawn from two interviews, lightly edited and condensed for clarity.

Let’s start with you. How did you end up doing this?

I studied at Carnegie Mellon University, on the Kigali campus, on a Mastercard Foundation scholarship. My master’s was in electrical and computer engineering, and my focus was machine learning. I applied four times before I got in, rejected three times, accepted on the fourth. 

Most of the AI products we see today aren’t really built for Africa, starting with the data they’re trained on. I wanted to change that, and to work on the problems that matter most here, in education, energy and agriculture. Coming back to Sierra Leone for this project felt like an opportunity to use my expertise where it’s needed most—helping my country, and the continent, navigate AI.

What did the ministry team actually want you to build?

My role was twofold: train the minister’s eight-person delivery team to use AI, then build a tool with them. Firstly, we identified a problem the team faced which AI could help with. The shift the delivery team were hoping for was to reduce the need to go through departmental heads every time they needed information, particularly when preparing donor reports or reports for partners. In many cases, the relevant information is either not shared with them, or is shared too late. As a result, staff end up contacting multiple people just to obtain a single piece of information, making the process inefficient and time-consuming.

So, they landed on an internal chatbot, meaning that a team member can ask it directly: how many students, how many teachers, what were the exam results—and get the answer immediately, for report writing or budgets.

Why WhatsApp, and why not build it from scratch?

Two practical reasons. First, connectivity in Sierra Leone is not reliable, and everybody already uses WhatsApp—so it’s fast, accessible and there’s nothing new to learn. Second, cost: building a tool from scratch would have run upwards of $100,000. When a platform that’s already widely used can do the job, building on top of it is a no-brainer.

WhatsApp is also heavily customisable. Behind the chatbot is a retrieval-augmented generation (RAG) pipeline: the documents are broken into pieces and stored so the system can search by meaning, not just keywords, then GPT-4 answers from what it retrieves, delivered through the WhatsApp business interface.

The tool was built on the ministry’s own data. Why is that significant—and what would have been sacrificed with synthetic data?

We ruled out synthetic data from the start. You cannot tell a ministry the number of teachers is approximately something; they need the exact figure to budget and plan, and a hallucinated number would actively mislead a decision. So the system only answers from a knowledge base I built out of verified documents, from the ministry, the Teacher Service Commission and a partner organisation. Every answer cites its source. If two sources disagree, the system won’t give a definitive answer until it’s verified. And there’s a survey built in: if an answer isn’t what a user needed, they rate it, and I can retrain so it improves.

What was the hardest part?

Data, without question. I’d say data is the fundamental task, because without it an AI system is of no use to anyone. There’s no central data system, and the quality and availability are a real problem. The ministry’s own website hadn’t been updated in about four years. So getting hold of usable, current data was the real work. I built on the foundations a partner had already gathered, statistics on schools, teachers and exams, and the Teacher Service Commission shared a lot after I did a demo on a visit. A big part of the job was just cleaning and structuring scanned PDFs and government documents into something a machine can actually read.

You keep coming back to data as the foundation. What do you mean by that?

That a tool like this is only ever as good as what feeds it, and it has to keep being fed. A sustainable approach needs three things: someone in the ministry who owns the data, the website updated regularly so recent data is published, and a pipeline to inject new records as they come in. There’s a wider pattern here too. Three of the six countries in the challenge independently chose retrieval tools like this, and I think that’s telling. The problem in our systems isn’t really collecting data, it’s that the data gets collected and then lost. You can’t retrieve it or make sense of it when you need it.

What would it take for a tool like this to last?

The foundations, before anything else. Computers that work, reliable internet, a place where the data lives centrally, and someone with the skills to maintain it. Beyond that, the policies: a data policy, and an AI policy. I’ve held that view since day one. The tool runs, and the team uses it most days, I can see it in the logs, so the demand is clearly there. Settling its long-term home, who hosts it and who keeps it fed after the first year, is the next step, and it depends on those foundations being in place.

How did working side by side—rather than having something handed to them—change how the ministry felt about AI?

I think it built curiosity and awareness, which matters. When the delivery team engaged, they were enthusiastic; they could see what the tool could do. By the end, some were asking for more training, which is a good sign. The appetite is there. The work now is to meet it with the basics underneath.

Looking back, would you do anything differently?

Honestly, I might have spent less time building the tool and more on the groundwork: digital transformation, helping leadership, helping the ministry set up its data policies. Other African countries are writing proper AI strategies for education right now, and that kind of foundation is what makes everything afterwards possible. If I were advising another expert starting an embedded role like this, I’d say: before you build anything, understand the infrastructure. Don’t assume there are computers, or internet connection, or that the data exists in a form you can use. Spend the first month just understanding what’s really there, then decide whether an AI tool is even the right next step.

Last one. If you had unlimited time and funding, what would you build?

My dream board wouldn’t start with a tool at all. It would start with the foundations: AI and data policies, and a decentralised, open data system the whole sector can build on, because you can’t do AI without data. Then I’d integrate this deeply with the Teacher Service Commission and the wider education system, and act as a dedicated AI expert guiding policy and training, so that ministers and teachers can use AI well, and ethically. The bigger goal behind all of it is AI that’s actually built for Africa, on our own data, working on the problems that matter most here.

Mohamed Alpha is the embedded AI expert for Sierra Leone in the Ministry of Education AI Challenge, an EdTech Hub programme.

Read the next interview >

< Back to all interviews