The Floor Did Not Exist
Attempts at this existed before us. They were not good enough for students raised on instant feeds and two-tap interfaces. Here is what we built instead, what 25,931 students did with it — and the thing that surprised us most, which turned out to be language, and the double tax it quietly charges every Ethiopian student.
Picture a Grade 12 student in Ethiopia, three weeks out from the exam that decides the rest of their life.
They want to practise on past papers. First they have to find them. That means scanned photocopies passed between friends, PDFs buried somewhere in a Telegram channel, pages missing, questions cut off at the margin, no reliable way to know which year it came from or whether the scan is even complete.
Say they find one. They work through a question and write down an answer.
Now what?
There is no answer key. There is no explanation. There is nobody to ask at eleven at night. They cannot find out whether they were right — which means the hour they just spent taught them nothing, or worse, quietly cemented a mistake.
That is the real starting condition. Not "students lack personalised mastery pathways." Students could not check their own answers.
Things existed. They were not good enough.
I want to be fair here: we were not the first people to notice this. There have been attempts — some of them well funded, some of them backed by institutions far larger than us.
They did not stick, and I do not think the reason was ambition. I think the reason was that they were built to a standard set by the people making them rather than the people using them.
The students we serve were born into speed. They have grown up with feeds that load instantly and interfaces that require no instruction. They will give you about two taps and a couple of seconds before they decide something is not worth the effort, and they are not being lazy when they do that — they are being efficient with the only resource they have, which is time, in a year when time is desperately short.
Against that standard, a resource that technically exists but takes ten minutes of hunting, three dead links and a pinch-to-zoom on a crooked scan is not a resource. It is a rumour of a resource.
So the bar was never "does this content exist somewhere." The bar was: can a tired seventeen-year-old, on a mid-range phone, on a bad connection, get from I do not understand this to now I do without ever wanting to give up?
That is the floor. And it is the floor that did not exist.
What we built
Several years of real ESSLCE past papers, cleaned up and put in one place.
Every question with its correct answer. Every question with a written explanation. And an AI you can argue with when the explanation does not land — ask it again, ask it another way, tell it you still do not understand and watch it come at the idea from a different angle. On the web, and inside Telegram, where students already spend their evenings.
Between April and June, 25,931 students signed up. 17,477 of them came through Telegram. Weekly active students went from 39 to 6,185 in eight weeks. No press. No launch event. No partnership with anybody.
The thing that surprised us most
We shipped a button that renders an explanation in Amharic. I expected it to be a nice-to-have for a minority of students.
It became one of the most used things we have ever built.
And it is worth being precise about why, because I had the feature wrong in my own head. It is not a translation button. A literal translation of a physics explanation is often worse than the English — the technical vocabulary has no natural Amharic equivalent, so you end up with stilted phrasing that is harder to read than what you started with.
What it actually does is re-explain the idea the way a friend sitting next to you would. It moves between Amharic and English mid-sentence, keeping the technical terms in English because that is how students genuinely talk about them, and carrying the reasoning in Amharic because that is the language they think in. It is not a translation. It is a classmate explaining.
Once we saw how heavily it was used, the reason became obvious, and it is something I now think about constantly.
Language is a double taxation on Ethiopian students. The exam is in English. The textbooks are in English. But for most students English is a second language they are still acquiring, so every single question charges them twice: once to decode the language, and again to understand the actual physics or biology underneath it. A student can fully understand the concept and still get the question wrong because the sentence defeated them first.
Nobody sees this in the results. A wrong answer looks identical whether the student did not know the subject or did not parse the question. The transcript just says they failed.
There is a reason this gets underestimated, and I think it is a specific one. A student who moves easily through English on social media, in captions, in song lyrics and film subtitles looks fluent — and by any conversational standard, they are. But the English of a physics explanation or a history passage is a different language in everything but name. Dense noun phrases. Passive constructions. Abstract nouns doing the work that verbs do in speech. A technical vocabulary where the meaning actually lives. Researchers in second-language learning have long separated the two: conversational fluency arrives within a year or two, while the academic register takes the better part of a decade. The first is routinely mistaken for the second — by teachers, by policymakers, and by students about themselves.
I want to be careful about how hard I push this, because there are two honest limits. Some of what looks like a language barrier is simply the subject: photosynthesis and opportunity cost are difficult words for native English speakers too, and there is no clean way to separate did not parse the sentence from did not know the idea. And I am not making an argument about what language schools ought to teach in. That is a far larger question than mine and other people are better placed to have it.
My claim is narrower. For students sitting an exam written in English, anything that lowers the cost of decoding hands them back attention to spend on the subject itself. That seems obviously worth taking seriously, and mostly it is not taken seriously at all.
Cutting that first tax — even partly, even imperfectly — gives a student back the attention they were spending on decoding and lets them spend it on the subject. I now think this is one of the highest-leverage things anybody can do for education in this country, and it is dramatically under-discussed.
We stumbled into it. The students taught it to us by using it.
What the students said
185 of them sent us 213 testimonials without being asked. Fifteen wrote in Amharic or Afaan Oromo. I have read every one.
"This app has done more for my education than I expected. It explained things my teachers couldn't, and it gave me confidence in myself. I truly think I have a chance of passing my university entrance exam because of it."
— Selam, Grade 12, Lafto Secondary School
"LOMI, you uplifted from me a lot of worries and tensions. Thank you very much and please continue the good deeds you are doing. A lot of people are in need of LOMI but they don't know."
— Abdulmejid, Grade 12
And this one, which matters to me more than any number in this post:
"It feels like someone is there to help instead of just giving answers, and that has made a real difference for me."
— Hawi, Grade 12, Green New Generation School
I wrote an essay in March arguing that a tool which simply hands over the answer has performed the lesson while the student sat and watched — that the difficulty is not a defect of learning, it is the learning. Hawi had not read it. She is telling me we got that part right, and I would trade a great deal of analytics for that sentence.
How this happened, since I think it is the useful part
I want to write this down, because if you are Ethiopian and building something, the how is the part worth having.
There is no company behind this in the way people assume. Lomi is built at night and at weekends. I have a full-time job. There is no telecom partnership, no ministry agreement, no distribution deal, no office. Others in this market have all of those things, and have had them for years.
We moved faster anyway, and I do not think it was talent. It was four decisions that anyone can copy.
Go where students already are. Not "download our app." Telegram is already open on the phone. 17,477 students never had to install anything or learn anything new to begin. Every step you put between a student and their first answer costs you most of them.
Do the unglamorous work. The single most valuable thing we did required no AI at all: find the past papers, clean them, structure them, attach the answers. It was tedious and slow and nobody had bothered.
Meet them in their own language — properly. Not a translation layer bolted on at the end. The way they actually speak.
Take cost off the table at the moment it decides everything. Every student this season used Lomi for nothing. That was deliberate. A student three weeks from the exam that determines their life should not be doing arithmetic about whether they can afford to study.
On that last one I want to be straight with you, because I would rather say it plainly than be vague and have it look like a retreat later: Lomi has to work as a business. Something that cannot sustain itself cannot keep serving anyone, and I have watched well-meaning projects in this country disappear for exactly that reason. So I am not going to stand here and promise that everything will be free forever, because I do not make promises I am not certain I can keep.
What I will commit to is the thing underneath it: a student's access to preparation should not depend on their family's ability to pay. There are many ways to honour that — and we will keep finding them.
Where my heart actually is
I wrote in March that mass schooling solved a genuinely hard problem — how to teach everybody — and that the price it paid was personal attention, because for two centuries you could have scale or attention but not both. That price is finally coming down. I believe that essay completely, and I want to spend years on what follows from it.
But more than nine in ten students fail the ESSLCE. In recent sittings the reported national pass rate has been in the single digits.
You cannot have a thoughtful conversation about pedagogical reform with a student whose one shot is three weeks away. Reforming a national education system takes a decade, and it takes ministries, teacher-training colleges, publishers, universities and examination boards. I am one person with a laptop.
So the order is: help them pass this exam, with the system exactly as it is. Then go and change the system.
That is not a compromise of the philosophy. It is the first step of it. A student who passes gets to go to university. A student who fails, in most cases, does not get a second chance at the life that follows. Everything I want to build afterwards is downstream of that morning in June.
What comes next
More students, earlier. The most common thing students told us, in a dozen different phrasings, was some version of I wish I had known about this sooner — not a complaint about the product, a complaint that they met it in June instead of March. That is the problem I most want to solve for the next cohort, and it is solvable.
More subjects, deeper content, and the parts of the exam we do not yet serve well.
And much more on language, because the students showed us that door and I do not think we have walked more than a few steps through it.
Twenty-six thousand students used something that did not exist four months ago, on the phones they already own, in the language they actually think in.
I know exactly where my heart is on this. It is with them, and I am not finished.