How I Turned My Interest in AI Into Real-World Impact in Uzbekistan
By Jaloliddin Musayev, Tashkent, September 2026
People usually ask me how I got into AI, and the honest answer is the SAT. I prepared for it alone for more than a year, without any course or tutor, and my only helper was an AI chatbot. I was asking it everything, from explaining grammar rules to deciding my study plan for the next month, and sometimes things that had nothing to do with the exam at all. In the end I got 1530. But I think the bigger thing I got from that year was a question. If AI helped one student in Tashkent this much, what happens if it becomes normal in whole Uzbekistan?
From that day I started learning it seriously. I want Uzbekistan to have its own AI one day. I know it sounds too big for a lyceum student, so for now I am doing it through a less exciting topic, air pollution data, and honestly the air is where I learned machine learning for real.
A little about me. I was born in Nishan district of Kashkadarya, my father is a driver and my mother is an engineer. Now I study in Tashkent at International House Tashkent Academic Lyceum, and this is my last year there.
The air
It started with Eco Pulse in September 2025. I wanted checking the air to be as normal as checking the weather, so I built a website that shows air quality and weather for 173 districts of Uzbekistan. It takes six pollutants and turns them into the EPA health categories, and it has a 48-hour trend and a 3-day forecast.
Around 1,200 people registered so far. It is not a big number, I know. For me the more important part is who is using it. 52 clinics use it to warn their patients on bad-air days, and more than 20 schools check it to decide if kids go outside for PE and recess or stay inside. I pitched every one of them myself.
There is one thing I did wrong on Eco Pulse at first. Many districts in Uzbekistan don’t have a monitoring station at all, so I trained a model to fill those gaps and put its predictions on the site. After some time I understood that these predictions are not measurements, but on the site they looked exactly like measurements. So I took them down. Later I rebuilt the model with satellite data and now it is back, but every number from it is labeled as an estimate.
Eco Pulse uses data from Uzhydromet’s stations. Uzhydromet is the national agency that runs weather and air-quality monitoring in Uzbekistan, so in August 2026 I used Eco Pulse to pitch myself to them, and they took me as an intern. Now I work with their station data, and I built a model that forecasts PM2.5 and PM10 three to four hours ahead for each station in Tashkent. They are testing it next to their own forecast now. Their forecast is still more accurate than mine, for PM2.5 their error is around 2-3 µg/m³ and mine is around 5. So I still have a lot to improve.
At the same time I am a research assistant in the air-pollution monitoring lab at TIIAME National Research University. Low-cost sensors matter for a country like ours because they are cheap, the problem is their readings drift. My job is to run sensors next to the lab’s reference instrument, log both of them, and check the data for gaps and patterns. I compared more than ten sensors like this so far. Once I noticed one sensor slowly moving away from the reference, and I brought it to my supervisor. The lab corrects these sensors and sends them to districts where there is no station, and now some of them work in Kashkadarya, Navoiy and Tashkent regions. The Kashkadarya ones mean a bit more to me. Kashkadarya is where I was born.
The biggest thing I made on the research side is the Eco Pulse Benchmark. I was looking for open station-level air data for Central Asia and I could not find any benchmark for our region, so I made one with two other people. It has 341,321 PM2.5 measurements from seven stations in six cities (Almaty, Ashgabat, Bishkek, Dushanbe, Khujand and Tashkent). We wrote seven quality rules before looking at the data, and every measurement had to pass them.
Those rules found some strange things. For example, two stations in Dushanbe that were listed separately were 99.99% the same over 33,462 hours, so it was one instrument counted twice. Satellite data was missing on 34.5% of station-days, and those days were 19% dirtier on average, which is bad news for models that lean on satellites. My own model came out a little ahead of the European CAMS forecast after bias correction, but the difference was not statistically significant, so I reported it as no clear winner. The data and code are open on Zenodo and GitHub.
Uzbek language models
This part is the closest to what I actually want to do. I am on the NeuronUz team, we build open language models for Uzbek and release them on Hugging Face. Uzbek has more than 30 million speakers, but for a long time there was basically no good AI in our language. I joined the team later than others. I train and fine-tune models, and my own part is Uzbek literature and culture. I collect that data, more than 50,000 examples for now, and train on it.
Right now our team’s 4B model has the same score as GPT-5.2 on UzLiB, a leaderboard that tests how well models know Uzbek, and it is above GPT-4o and GPT-5 there. That is the team’s result, I am one part of it. But when I say I want Uzbekistan to have its own AI, this is kind of what I mean.
Other things I do
Not everything is code. I co-founded Blue Zarafshon, which started as a small cleanup in our neighborhood and now is a youth eco-movement in four or more regions. More than 500 volunteers came to our events, and I mean people who really came, we have the lists and photos. We planted more than 120 trees so far. We also recycle more than 100 kg of paper a year, and our lessons reached more than 20 schools. Our events were an eco version of Zakovat, an eco debate, plogging (jogging and picking up trash at the same time) and an eco quest with posters. UNDP and Eco Heroes supported us, and UzA wrote about us.
In August 2026 I organized TexnoNova, a youth hackathon in Fergana around one question, how can we reduce air pollution in cities. We ran it together with IT Park’s Fergana branch and Startup Garage. More than 80 people came, and more than 15 teams submitted projects in three tracks. We had experts from IT Park and from the government, and about 10 million soums in sponsor vouchers. One of the winning teams kept working on their project after it ended.
I also started AI Community of Uzbekistan (ACU), and I co-founded Eengineer.net, where I do marketing and finances. Together they have more than 300 student members. I think most students here don’t know how to reach researchers abroad, so I started writing to them myself. I cold-emailed more than 200 professors. Most of them said no or didn’t answer, which is normal. More than 20 said yes, including scientists from Mount Sinai in New York and Leipzig University. Through ACU we connected more than 50 students, and it turned into a research collaboration, an internship and mentoring that still continues.
Because the SAT is where everything started for me, I co-founded SAT TOP (sattop.uz), an SAT prep platform, and I built the platform. I also teach SAT. It grew from one-on-one help I was giving to students who couldn’t pay for courses, and one of them went from zero to 1400. Now our Telegram community has more than 16,000 members.
At my lyceum I am chief publisher and team manager of our school newspaper, the Main Press Department, we are six people. The paper never wrote about air quality before, so I started that, and more than 400 students read our first air-quality piece. We printed at least three editions so far, around 100 copies each.
Before Uzhydromet I did two other internships. At Uztelecom I was a software engineering intern and built four internal apps that engineers there use. At the Youth Affairs Agency I prepared more than 50 official documents for the Cabinet of Ministers and the Ministry of Education, and taught four other interns Excel and Word. On top of that, I am building Qayta now, a compliance tool for plastic recycling companies, and 10 companies already use it.
Swimming
Outside of all this I swim. I train and compete at Tosh Kuch Swimming Club in Tashkent, and I have some 1st, 2nd and 3rd places from regional championships. This summer I came 3rd of nine in my heat at the Born2Swim challenge in Tashkent, where swimmers were from Uzbekistan, Kazakhstan and Kyrgyzstan. I also finished the 2.2 km open-water swim at Charvak, which gives you the Charvakman title. At the same club I am an assistant coach for 30 kids aged 5 to 12, and seven of the kids I coach won their first awards at a local tournament.
I am also genuinely into touch typing. I type around 100 words per minute now, which is top 8%, and I am training for a competition.
What I want to do
If a younger student reads this, my one advice is write to people. Most of my emails went nowhere and that is fine, 20 yeses were enough to connect more than 50 students. Also, not having money for courses does not mean you can’t prepare. I prepared for the SAT without one.
The AI use cases that excite me the most are the boring ones nobody talks about. In Uzbekistan a lot of paperwork is still done by hand, which wastes time and sometimes creates corruption. Our farmers lose water and harvest because decisions are made by guessing, AI can help there too. A student who prepares alone like I did could have a mentor that speaks Uzbek. These are not shiny use cases like image generators, but in my opinion they change real lives.
I am applying to universities now for fall 2027. I want to learn machine learning properly, with real fundamentals, because right now I am building things half-blind. After that my plan is to work at a frontier AI lab for some years, and then build my own company and my own language model. That is the plan at least, we will see how it goes.
LINKS
Eco Pulse: eco-pulsy-5of6.onrender.com
Eco Pulse Benchmark data: doi.org/10.5281/zenodo.21930669
Benchmark code: github.com/jalencik/eco-pulse-benchmark
NeuronUz: huggingface.co/NeuronUz
UzLiB leaderboard: github.com/tahrirchi/uzlib
SAT TOP: sattop.uz
My GitHub: github.com/jalencik
Email: jaloliddin2009applicant@gmail.com
रिपोर्ट : प्राची गुप्ता
