Why ControlAI¶
The pain: the owner can't see what's really happening in lessons¶
A learning center sells one thing — teaching quality. Yet that is precisely what the owner can barely control. They can't sit in on every lesson, and there are dozens of lessons a day.
Here's what happens in practice:
- The cameras are there, but no one watches them. Recordings pile up "just in case" — people only go back to them once a scandal has already broken out.
- Problems surface too late. That a teacher talks half the lesson themselves, asks no questions, or is regularly late — the owner finds out months later, once students have already started leaving.
- There's no objective picture of teachers. Who actually carries their weight and who merely goes through the motions of "running a lesson" — you can only tell by eye and from complaints.
- Students leave quietly. A parent doesn't write "we didn't like it" — they simply don't renew the subscription. And the reason was often visible in the lessons all along.
Manual oversight doesn't solve this: putting a person on watching every lesson is far too expensive, and even then they'd only manage to check a couple of percent.
What ControlAI covers¶
ControlAI makes the invisible visible, for every lesson:
- Transparency. The owner sees what happened in any lesson without re-watching the recordings.
- Early warning. A weak lesson, a spike in late arrivals, a drop in activity, a group on the verge of churning — all of it is visible right away, not after the fact.
- Objective evaluation of teachers. The same metrics for everyone, week-by-week trends rather than gut feelings.
- Method compliance. Each lesson is checked against the rubric and the curriculum of your specific center.
- Student retention. Problem groups are visible in advance — there's time to step in before a student leaves.
And all of this comes with no extra load on staff: the system works on its own, and people only look at the finished result and make decisions.
Why not "just upload the video to any AI"¶
A fair question: "There are smart AIs now — why doesn't the center just take ChatGPT or a similar service, upload a lesson recording, and ask how it went?"
It doesn't work that way — and here's why:
| "Upload the video to an AI" | ControlAI | |
|---|---|---|
| Volume | Manually, one clip at a time. And that's ~120 hours of recording per day per center | Automatically pulls and analyzes all lessons in all classrooms every day |
| Cost | Paying a big AI for every minute of video — unaffordable at the volume of lessons | Most of the metrics are computed by ordinary algorithms, not heavy AI; the expensive AI is only needed where nothing else will do — to write the report text from ready-made numbers and to check the lesson against the center's rubric |
| Memory | One conversation, and it's all forgotten: no history, no trends, no comparison | Remembers every lesson, accumulates history per teacher and per group, builds trends over time |
| Children's data | Video and voices of minors are sent to someone else's servers abroad | Recordings are processed on our servers in Uzbekistan; raw video/audio does not go to external AIs |
| Result | A vague summary along the lines of "overall not bad" | Precise, comparable figures: talk-time 68%, 12 questions, 3 late arrivals, English 40% |
| Who spoke | Produces a solid block of text — it can't reliably say where the teacher is and where a student is | Specifically separates the voices and knows who the teacher is — so it accurately counts who spoke how much |
| Your methodology | The curriculum can be pasted into the prompt — but by hand and from scratch for every lesson | The center's rubric and curriculum are set once and apply themselves to every lesson, with cumulative tracking of topics covered |
| Is it a system? | A one-off experiment in a chat | Schedule, roles and access, personal dashboards, consents, reliable processing, history for everyone |
The key difference: ControlAI remembers and accumulates, a chat does not¶
An external chat AI does not collect the lessons itself and does not store them as a structured, comparable history. Noticing that a teacher has been slipping three weeks in a row, comparing groups against one another, showing trends over time — it cannot do without someone manually uploading and re-explaining every lesson from scratch.
ControlAI is the opposite — a permanent system with memory. From this grow capabilities that a one-off chat simply doesn't have:
- History for everyone. All lessons of every teacher and every group are stored and linked together — instead of being lost after a single answer.
- Trends and dynamics. A teacher's rating over 8 weeks, a group's progress, a rise or fall in activity — you see movement over time, not a single snapshot.
- Comparison. Teacher against teacher, branch against branch, group against the center's average.
- Groundwork for the future: it gets smarter. The center's accumulated data will allow us to keep improving recognition — including Uzbek-Russian speech, which ordinary models understand poorly (see the Roadmap). This is what can't be replicated by "uploading to a chat."
- It works on its own. It pulls recordings on schedule, computes, and flags problems itself — there's no need to upload anything or ask by hand.
- It knows the context. Who the teacher is, which group it is, which topic is on the plan for today — so it checks not "an abstract clip" but your specific lesson against your curriculum.
To put it simply — a standalone AI can "have a chat about one clip." But the center needs something else: for every lesson, every day, to be turned automatically into uniform, precise, comparable metrics that accumulate into history and add up to trends — economical at scale, safe for children's data, and aware of the methodology of that specific center. That is ControlAI.
The topic of data protection is only outlined here — we cover it in detail (consents, storage in Uzbekistan, deletion of recordings) in the chapter "Privacy and trust".
The key takeaway from this chapter: the center's pain is that the owner can't see what happens in lessons and learns about problems too late. ControlAI makes every lesson transparent and measurable. And "just upload the video to an AI" doesn't solve this task: it's not about one clip, but about volume, cost, children's privacy, precise figures, and the methodology of the specific center.
→ Next: 03. The journey of one lesson — we'll walk through it end to end: from the start of the lesson to a finished report in the admin panel.