The journey so far

Near the end of that second term, a teacher sent me a link to an AI Learning Hackathon, open to all of Rwanda, with only ten schools selected. We were chosen. Our team of three, HAGENIMANA Samuel, ISHIMWE Eric, and I, left our end-of-term exam to compete in Kigali, alongside students from RCA — the school I'd once hoped to attend.

Our pitch, on day one, revealed something unsettling: our idea and RCA's team's idea were nearly identical. And unlike us, they'd arrived with a finished, production-ready product. We had arrived with only an idea, believing incorrectly that building was part of the competition itself. We spent the final two days building without sleep — never once using the hotel beds we'd been given.

We built GuideMe, a personalized learning platform, with me handling the Flask and Python backend. On the final day, exhausted and sick from three sleepless nights, our project's core dependency — a ChatGPT API key — hit a rate limit during rehearsal, minutes before we were due to present. There was no way to recover it in time. We presented anyway, and lost any real chance at placing.

RCA's team took second place. First place went to a team from a school with no software development program at all — their prototype, a simple, solid tool to help blind users operate a computer, had been built the night before with help from HAGENIMANA Samuel himself.

Months later, working through an OpenCV project, Samuel came across something that put the whole competition in a different light: RCA's polished, "production-ready" entry hadn't been built from scratch during the event at all — it was a customized template from Soul Machine, a company known for hyper-realistic, emotionally intelligent digital humans, styled to look like an original product. We had no idea at the time, and it doesn't undo anything we learned there. We'd written every line of our own code that weekend, and it still wasn't enough to place — which taught me something more useful than a clean loss would have. It's a stranger lesson to sit with once you learn the playing field wasn't quite what it seemed, but it's the same lesson regardless.

President Paul Kagame, who helped award the prizes that day, announced that every participant would receive a new laptop. I'm still using it today.

When the competition ended, we came back just in time for the internship every L4 student takes — two weeks of the normal holiday combined with two weeks provided by RTB, forming one full month. Before it began, I pulled together the five of us who could make it — HAGENIMANA Samuel, RUKUNDO El-shaddai, MUHOZA Elan Clever, RUKUNDO Jean Baptiste, and I — and made a decision: instead of treating that month as a holiday, we'd turn it into our own internship. We rented two houses in Gasabo, Kigali — one to live in, one that became our office — paid a month upfront, and built out of it what an actual company internship might have looked like, entirely on our own terms. That's where the name EduBridge was actually born. ISHIMWE Eric wasn't part of this stretch — he was in Uganda at the time.

IGABE MUSANGAMFURA Lydivine and ISHIMWE Romain weren't living in Gasabo with the rest of us, but they were part of EduBridge from day one all the same, shaping the idea with us throughout even while working from outside that house. All seven of us stood together for the actual presentation, taken to RTB by our teacher — still mostly a concept at that stage, with a partial prototype rather than anything that could genuinely predict a student dropping out. RTB never called the idea bad; they just weren't confident we were the right team to carry it forward, and said they'd follow up. They never did.

When the third term started, I kept working on it on my own, refining it until it became something real — an actual model capable of predicting a student's likelihood of dropping out, built out from that unfinished internship idea. Years later, after we'd all graduated, RUKUNDO El-shaddai went on to found LinkFy Connect.

That holiday, heading into L5, I taught myself Django from another YouTube course. Not long after, I built and sold my first real client project — a complete school website, for 500,000 RWF. It's still live today, still used by that school.

In L5, with graduation only months away, I watched classmates plan to start web agencies — building sites for clients. I knew I wanted something else: to solve a real problem, starting from my own community, the way EduBridge had tried to.

At the time, AI still struggled with Kinyarwanda. I briefly considered building a Kinyarwanda-language model myself, until I noticed the existing models were improving at that faster than I could catch up to. So I looked past language, and found something underneath it: speaking a language fluently isn't the same as understanding how a place actually works. My own father struggled to get useful advice from AI — not because it couldn't speak to him, but because its answers were generic, built for no particular context, while his life was built entirely from one.

The father story was one thread, but not the only one. The more I looked, the more I found the same pattern repeating — in healthcare, agriculture, business, government, in ordinary daily life. Something as simple as a birth certificate should exist as retrievable data, yet still usually means an in-person trip to the local umudugudu office. A patient needing urgent care doesn't always have a fast way to reach one. Calling a service center for something routine can mean two hours on hold before a person answers. None of that is a technology problem in the way people usually mean it — it's a systems problem, the kind a general-purpose AI, however fluent, was never built to notice. Those were the moments that shaped what ATAS eventually became: not one product for one use case, but infrastructure meant to sit underneath problems like these, wherever they show up.

I chose the name deliberately. The "Alliance" in ATAS was never incidental — I knew from the beginning that whatever I built would eventually mean working alongside others, even if I started alone. I remember asking a classmate to help me get the spelling exactly right, half-worried people would misread it as "ATASA."

When I graduated, I kept the promise I'd made to myself and continued building what I'd started in L5 — AcademiaPlus — the project that, alongside ATAS, still occupies most of my time today.