A day in the life¶
To make the product fully concrete, let's walk through a few real-life scenarios — how different people at the university actually use ControlAI day to day. Names and numbers are fictional.
Morning at the quality-control office¶
08:40. The head of the quality-control office (ichki ta'lim sifati nazorati bo'limi) opens the panel. No need to call around the dean's offices or request written reports — yesterday's full picture across every faculty is already on the dashboard: of 312 scheduled class sessions, 303 were held, 9 didn't happen, and 6 of the ones held started more than 10 minutes late.
At the top, in the "Needs attention" feed, there are three lines. The first, in red: "Department N — 7 class sessions not held this week, all from two instructors." He expands it — every session is visible: room, time, a "session not held" mark taken from the recording, not from someone's word. He forwards the breakdown to that faculty's dean with a short note: look into it and report back.
The second, in yellow, is the review queue: for yesterday's sessions, ControlAI has already filled in the observation forms, and 14 of them are flagged "needs a human look." That volume used to take a committee a week — now it's an hour of careful checking before publishing. The third, in green — "Faculty M — best week: 100% of sessions held, highest average instructor rating" — he adds it to the materials for the upcoming rectorate meeting.
The dean's day¶
09:00. The dean starts the day with yesterday's summary for the faculty. Coverage is normal: of 84 scheduled sessions, recordings were obtained for 83 — one room's camera is offline, and the task is already with the technical service; 78 sessions have been processed, the rest are in the queue. One session is marked "not held": there's a recording, but the 2nd-year group didn't show up for the practical class — the system flagged this automatically, and that session isn't included in the billed volume.
Next he opens the department comparison view: for one department, students' talk share in seminars is noticeably lower than the faculty average, and the monthly trend is heading down. The dean doesn't guess — he forwards the breakdown to the head of that department with a short note. By lunch, tomorrow's schedule has been checked and there are no conflicts.
The head of department's week¶
Tuesday. The head of department opens the methodology review section. Yesterday's session from a junior instructor is flagged — ControlAI has already filled in the university's own lesson observation form: in the past, a committee filled out that kind of form only once or twice a year, during an open (demonstration) lesson (ochiq dars) — now there's one for every class session.
He goes through the criteria one by one. The AI has confidently filled in most of them, with quotes from the class — he simply confirms those. A couple of items are flagged "needs a human look" — he spends his attention there: opens that moment in the recording, watches it, and enters his own verdict. The picture is clear: the instructor talked for almost the entire session himself (his talk share is 88%) and asked only 6 questions in 80 minutes — a classic first-year mistake.
On one criterion, the AI underrated "clarity of explanation" — the head of department corrects it, and that correction isn't wasted: next time, the system will score a similar case more accurately. Instead of calling the instructor in for a dressing-down, he brings in the mentor for junior instructors, and together they map out a plan: more questions to the audience, pair work in the second half of the class.
After class — the instructor¶
11:50. The class has ended. The report is usually ready within an hour (by the next morning if the internet connection is limited), and by lunchtime the instructor opens his dashboard — the class breakdown is already there, in his group's language. He used to have no idea, in numbers, how his classes were going — now he sees: students talked 34% of the time, 14 questions were asked, his rating is 6.8 and has been climbing for three weeks running.
His goal is to raise students' talk share in seminars to 40%. The progress bar is almost full. A tip suggests a bit more small-group discussion in the second half of the class — he tries it in his next session, and a week later sees in his trend that the number has gone up. For him, this isn't "oversight from above" — it's personal statistics and a clear game of improvement: the rating is an internal development tool, not a replacement for official certification, and instead of a subjective committee visit once or twice a year, he has a fair, consistent measure of every single class session.
Once a month — the rectorate¶
The rector doesn't sit in the panel — and that's a deliberate part of the product: the system runs through the hands of the relevant offices, and leadership gets the finished result. The vice-rector for academic affairs reviews the university-wide summary once a week, and once a month brings the outcome to the rectorate meeting: faculty comparisons, sessions that weren't held and why, rating trends, the best departments. The discussion runs on the same numbers in front of everyone, not on impressions or memos. And the evidence base for accreditation and national rankings builds up on its own — no one needs to open the panel for that.
Key takeaway: in real life, ControlAI saves time and removes the guesswork: the quality-control office does in an hour what used to take a committee a week; the dean keeps their faculty and department comparisons under control; the head of department reviews every session from a junior instructor instead of one or two open lessons a year; the instructor grows against clear goals; and the rectorate gets the finished outcome at the meeting without spending time in the panel.
→ Next: 12. What the university needs — what it takes to make all of this work.