My name is Aidin Biibosunov. I teach mathematics in a rural school in Kyrgyzstan’s Chuy region as a jaratman (fellow) of the Teach for Kyrgyzstan Initiative, known locally as Zamanbap Mugalim (“Contemporary Teacher”). The organization is an education leadership program and a member of the Teach For All network. Together with Zamanbap Mugalim, I am developing AIsuluu, a system that grades handwritten math homework and sends each student personal, constructive feedback.
The problem that starts after the bell
The hardest part of my job is not explaining fractions or equations. It is what happens after class.
In many rural Kyrgyz schools, one mathematics teacher teaches 40 to 50 students. Checking that many handwritten notebooks each evening leaves little room for teaching. Most children get a note informing them of their score, or nothing at all. They rarely learn where they went wrong or how to fix it.
That feedback gap hurts everyone, and it hits some students harder. In many rural communities, cultural expectations encourage girls to stay quiet and deferential in class. They are less likely to ask questions or seek clarification. When a teacher’s resources and attention are limited, students who speak less often receive less guidance.
Nationally, girls often score higher on exams than boys, yet choose technical fields less often. According to CABAR.asia, citing National Statistical Committee data from 2022, girls scored 123.3 on the nationwide Main Test compared with 115.4 for boys, but only 23 percent of women applicants chose technical specialties versus 38 percent of men.
Without individual guidance, school attendance can become hollow. Stronger analytical skills also help young people evaluate information and take part in public life. Closing the feedback gap is both an education problem and an inclusion problem.
What we set out to do
AIsuluu has three clear goals. First, give every student who submits homework equally detailed feedback, regardless of gender or classroom participation. Second, give teachers back hours each week so they can teach rather than only grade papers. Third, show that this tool can work within the structures rural schools already have, then publish the code for free so that others can adapt it.
Our direct beneficiaries are rural math students. We are piloting the system in my classroom at Novopokrovka Secondary School No. 3, with a plan to reach about 150 students across three Teach for Kyrgyzstan partner schools in the Chuy region. Teachers and fellows who use the tool daily are secondary beneficiaries.
How it works in the classroom
Fundamentally, teachers run their own classrooms. They create each assignment, upload a reference solution, and can choose practice materials or next steps for students who need more work on a topic. The AI tool does not decide what to teach; it helps carry out the teacher’s own lesson plan.
The student workflow is intentionally simple. A student finishes their homework, photographs the worksheets with a smartphone, and uploads them through a form. A vision-capable language model reads the handwriting. A second model compares the work with the teacher’s reference solution, scores correctness, process, and presentation, and provides feedback in plain language: Instead of just “right” or “wrong,” students will get detailed feedback, such as “you forgot to change the sign when moving the term in problem 3.” The students receive the results in their email inbox in the language of instruction, whether Kyrgyz or Russian. Teachers can review borderline cases before feedback is final. The system is an assistant, not a replacement.
What has worked so far
The strongest step so far has been the living pilot in a real classroom, not a lab demo. Watching the system handle messy handwriting, award partial credit, and return specific next steps convinced skeptical colleagues that this was more than a gimmicky gadget. I also presented an academic paper on AIsuluu at the KGIPI Conference in January 2026, which helped frame the work as pedagogy, not novelty.
Students’ reactions are still early, but a pattern is emerging: When feedback arrives within minutes instead of days, they reopen their notebooks. Teachers notice something even more promising. Girls who rarely ask for help still receive the same depth of comment as the most assertive boy in the room. That equality is built into the design.
What comes next
By September 2026, we aim to run AIsuluu across three partner schools, train teachers alongside fellows, ensure feedback works in both languages of instruction, and publish the full codebase on GitHub under an open license. We will measure submissions processed, time saved for teachers, and shifts in proficiency and confidence broken down by gender.
Sustainability is part of the design: Open-source code, IUCA hosting, and a path toward open-weight models to reduce API costs. Teach For Kyrgyzstan’s network of partner schools and fellows offers a ready path to scale. I have already shared AIsuluu with the Teach For Kyrgyzstan community, and interest in replication is growing.
What we have learned
I built AIsuluu because of something I met in my own classroom. Too many notebooks, too little time for real feedback. I already knew how to work with AI and code, and the smartphones were already in students’ hands. So I stopped waiting for a school reform and tried to solve the problem with the skills and tools I had.
Along the way, I learned to lower the ambition and raise the frequency. Better a simple comment every week than a perfect system nobody uses. Students do not need a long essay from the AI. They need one clear note: Where they went wrong, and what to try next. Short and useful beats impressive and vague.
The classroom keeps correcting the project which is how it should be. Once the habit exists, improving the tool is easy. Building the habit is the real work.
