Fastagger, a Kenya-based AI-as-a-service Startup, Sees Growth Opportunities In Geospatial Data

According to the Government Artificial Intelligence Readiness Index 2019, a research conducted by Oxford Insights and the International Development Research Centre, lack of systemic study on artificial intelligence (AI) and associated dearth of contextual data on Africa are significant challenges to the progress of AI applications in Africa.

“This means that applications of AI developed in other regions will likely lack contextual relevance, particularly in regards to cultural and infrastructural factors, and will not be fit for purpose in Africa. For example, a lorry in the United States or Europe encounters substantially different challenges to a lorry in most of Africa. A self-driving lorry developed for the roads of developed countries is unlikely to be successful on the roads of developing countries without substantial adaptation,” the report says in its synopsis on Africa.

To address this challenge, Kenya-based Fastagger, a startup offering artificial intelligence as a service (AlaaS), has set out to accelerate the adoption of AI applications in Africa by providing contextual data on the continent and helping entities develop bespoke AI application use cases.

The startup’s AlaaS platform provides image annotation services to AI-driven businesses across multiple sectors, including healthcare, energy, financial services, industrial and agriculture. 

Fastagger is among the ten African startups selected for the 2020 cohort of the Africa4Future aerospace acceleration program by Airbus and GIZ, and implemented by CCHub. As part of the program, the startup has been focused on developing an AI-based, satellite-enabled solution for solar infrastructure companies to efficiently map potential markets according to detailed parameters such as rooftop count.

We recently had an interview with the company’s CEO and co-founder Mutembei Kariuki as part of Space in Africa’s special series on the ten finalists for 2020 Africa4Future program. 

Read the full interview transcript.


I understand you have an excellent team of co-founders: Jude Mwenda as a CTO and Stephanie Njerenga as COO. What is the inspiration behind founding Fastagger, and how did you put together the fantastic team?

Incidentally, if I think back to where the ideas came from, I remember way back to my childhood because my dad introduced me to computers. He bought second-hand computers which were intended for disposal by one of the companies in Kenya. I think the PC was running on Windows 95, but it had a program called “Eliza”. Eliza was one of the first AI chatbots ever developed in the 60s. I used to interact a lot with Eliza, and that developed my interest in computers and IT. 

Fast forward to many years later. I spent time in Japan, where I learnt about opportunities and trends in the AI space, including data transcription and elliptic speech. I thought AI would create disruption globally and affect many jobs, but there’s also a potential for each to create millions of jobs. 

My co-founder Jude and I have known each other for several years; I think ten years and we were working on our previous startup, working on data and seeing how technology was going to impact the continent. That was about 2016/17 before he left for MIT. But we kept in touch. He ended up starting his PhD in AI at Georgia Tech, while I was part of a startup program in Kenya. At some point, I had the idea of the startup, and he was like, yeah, this is connected to what I’m doing. That is how we got involved.

Stephanie, my other co-founder, and I have known each other for a couple of years. I used to work for GIZ and with the startup ecosystem in Kenya. At the time, she was doing management consulting and was working to support tech startups in Kenya before travelling to work in India and then Spain for her MSc in Big Data and Analytics. With her background in actuarial science and having worked for one of the first machine learning startups in Kenya, we started having a conversation. She picked interest when I pitched the idea, and we decided to come up with a formula of how we would start. That is how we started with data labelling, because any AI learning through machine learning, which is called supervised learning, has to be taught by lots of labeled data. 

We saw that opportunity because we are not only thinking about the idea of making money but also how we can democratize AI. And how we can create opportunities for young people on the continent because AI is going to be very disruptive.

Great! When did you finally launch the startup?

We started last year and are very young. Just generally also because a lot of better evolutions of AI have happened during the last five years. Much suitable ML and deep learning models have been developed during this time. We saw that the right time to do this is now and we see different opportunities.

What is the size of your team at the moment?

There are three of us who are the co-founders. Then we have two team leaders working with about 100 young people who are working for us as independent workers doing the data labelling. We expand these teams based on the pipeline of tasks that we have. And we have some partnerships to help us to do this.

I see you mentioned connecting Africans to the future of work and that you are helping up to 50 Africans who are part of your developer network. Can you talk about that? What is your relationship with these young Africans?

We mainly target young African people from low-income space who have some digital skills and do have access to the internet and computers, and we work with them to do the labelling of data. In the AI space, it is like teaching a small child basic language such as what is an apple, what is a car, what is a person? You need to show them images in a textbook where someone has written man, woman and the picture of a man or a woman. This approach is the same way machines learn. So imagine now you need to create this curriculum or you need to create this many books. And then you need to show this child many pictures and do it every day. That is the same way you train a machine, especially to recognize what is an apple, a man, a dog, a cat. 

To teach these machines, you need to feed them with hundreds of thousands of such images. The reality is that you need people, humans to do that labelling to get the pictures and to do that labelling. The role has created an opportunity for young unemployed people with digital skills on the continent to work and earn a bit of an income. That is how we are connecting young people to the future of work so that during this process, they’re learning what AI is. As we are designing and building more models for different use cases in Africa, we are also partnering with other organizations to educate and upskill them to do more complicated technical work.

That’s fantastic! For these young people, what specific skills are you looking out for, and what is the geographical spread of the recruitment?

Our vision is to scale up across the continent in the next couple of years to impact up to a million young people in the future. So right now, we’ve started small, of course. As for skills, we are working with people who have basic computer skills. As long as you know how to, open your browser, use a mouse, access the internet to know, log on to a platform, then we give you the training on how to perform specific tasks. So these are the fundamental skills we require from candidates in the case of data labelling. As we go to more complex things of building models, we shall look for people who have experience in software development. And those are the ones who over the coming years, we are going to upskill to do more technical work.

How is the startup funded? Have you raised any external investment yet?

Raising external investment has been quite a challenge for us, especially now because of COVID-19. We had difficulty convincing people that AI can be a profitable venture in Africa. We have had investors just tell us to our face that AI cannot be done in Africa because all the AI people are in Silicon Valley. Sometimes, it might just be the implicit bias of investors who do not expect Africans can do some of this work. However, we have also been able to find some angel investors from whom we raised angel funding. But then we are more focused on driving revenue. So with that small angel investment that we have right now, we are focusing on getting revenues. We have been able to drive a bit of revenue and are working on some commercial projects in the pipeline.

You mentioned there are some projects in the pipeline. Does that include the project that targets the energy sector in Africa? I understand that might be one of the reasons you got selected for the Africa4future acceleration program. Can you talk about it? Where are you on the project?

Regarding this particular project, I have a bit of a background in urban development using GIS. Likewise, my co-founders – Jude has much experience in GIS, and Stephanie has expansive knowledge in machine learning and algorithms. 

As part of Africa4Future program, we were looking at what problems on the continent can be solved using satellite remote sensing data. One of the companies that we approached is a solar infrastructure company. They set up mini-grids, and we talked to their GIS engineer who confirmed that they faced a significant challenge in assessing the markets where they are growing. They access satellite images of potential markets and manually count the number of buildings and rooftops in a particular area for installing mini-grids and solar panels. Due to the time constraint and inefficiency in doing this, they tried to develop algorithms to automate the process but could not train the data. 

We saw this as an opportunity to solve a problem because there are about 600 million people in Sub-Saharan Africa, who don’t have access to electricity. We saw this as a problem that can be solved using remote sensing data and AI. We started working on a solution and, so far, the model has about 50% accuracy of being able to detect rooftops and solar panels. We are working to build that up to a good enough accuracy using the images that we have access to through the Airbus UP42 partnership.

When will the solution be ready for commercial clients?

It will take a while because machine learning and AI are not very fast processes. In about six months, we should be able to deploy it with a good enough accuracy to be used successfully by our clients.

Do you have other services or products that already have commercial clients?

Yes, the commercial service that we have is the data labelling. We have clients that leverage our data labelling services. Also, we have corporate clients who want machine learning algorithms developed for them for solving specific challenges in their businesses. We are currently presenting use cases to some clients for two offerings of our products: data labelling and the AI-as-a-service.

What exactly does the AI-as-a-service do for customers?

So imagine using email, instead of owning your server, and trying to build your email infrastructure, you just create a Gmail account and let Google manage backend services for you. AI-as-a-service is similar in the same way.

If you have a use case, let’s say you need to be doing identity authentication, using face ID. So you want to identify if a specific person has the right access to a place, and you don’t have the technology to do that or the infrastructure. For us, we build the AI that does the facial identification, and we provide you with access to it. You pay a subscription fee to use the service. You don’t need to build or own the algorithm yourself.

That is one example of the subscription model. The other scenario is when we identify business challenges a company is having, we then propose to build for them an A.I-based solution. They pay us over time for using the AI because they might not have the infrastructure to build it. 

Artificial intelligence and machine learning and most of the cutting edge technologies are relatively early in Africa. Do you see considerate market demand, and do you think the market is inspiring your product development?

Yes. There is a market readiness report that was done by Microsoft for Nigeria, Kenya, South Africa, Egypt, and Morocco exploring corporations and application use cases in these countries. There’s much movement towards the use of AI in Africa generally. Especially, now because of the disruptions that have been caused by COVID. People are going to start speeding up digital transformations in their organizations. They are going to start thinking about how they can use AI.

After the Africa4Future showcase, we gained more traction with many companies approaching us to learn how our AI can improve their business processes, especially in the financial and health sectors.

Right now, most of the labelling work we are doing is for clients who are in the US, Asia and Europe. We see that the market is still growing. However, we want to be the ones who are going to help the market to grow here in Africa, because AI is very much like electricity, for you to generate electricity, you need to be near the generation source. AI is very contextual and requires contextual data. We cannot benefit much from AI developed outside of the continent. We need to develop AI-based on the specific needs we have in each of the 54 African countries. 

Interestingly, you mentioned contextual data for AI, especially in Africa, where there is often a scarcity of data. You said you see growing demand from Asia, Europe and the US? Are these demands for contextual data in Africa?

No. The demand is for data labelling for their use cases. But we have talked to some researchers who are looking to develop models for Africa. For example, in health care for skin cancer detection. In this case, the data they use is from Caucasian skin, and that cannot work appropriately in Africa for a majority of the population. We have had corporations with such interest in contextual data from Africa approach us. They have their own AI and approached us to help them to do the labelling of hundreds of thousands of images and data that they have.

Do you see further intersections between satellite and space technologies with AI in terms of application use cases?


Yes, absolutely. There are so many use cases. Satellite data can be used in the real estate space, retail, agriculture, financial markets, forestry, security and maritime surveillance and many more application areas. 

What is your experience with the Africa4future Aerospace Acceleration program? What has changed in your business model as a result of what you learned from the program?

We were not focused on the use of satellite imagery and the opportunities there before we joined the accelerator. So the accelerator opportunity has been fantastic in terms of opening us up to other possibilities of using satellite data which we thought was limited to agriculture. It has opened our minds as well in terms of the technical expertise and connection to Airbus, an industry leader. We see Airbus as a potential client and partner, and more people have come to trust our reputation because we participated in the acceleration program.