What is ControlAI¶

In a nutshell¶

ControlAI is a lesson analytics system for learning centers.

Cameras already hang in every classroom. But almost no one watches the footage — it is a "dead archive" that people return to only when something has gone wrong. ControlAI takes that footage and turns every lesson into a clear report: who talked and how much, how many questions there were, how many students showed up and how many were late — and whether the lesson followed the center's methodology. And over time it will also notice phone distractions and rude language (that part is on the roadmap).

As a rule, no new devices need to be installed. ControlAI works with what is already in the classroom and analyzes every lesson — not one in a hundred, as with manual review.

A little more detail¶

Imagine you have a very attentive assistant who can:

"listen to" a lesson and figure out who spoke and how much (the teacher or the students), how many questions and pauses there were, which language was used — and, over time (roadmap), also notice praise or rude words;
"watch" a lesson and figure out how many students were in the classroom and how many were late — and, over time (roadmap), also notice phone distractions or a "head on the desk";
check the lesson against the center's methodology — whether the teacher covered the planned topic and did so the way the center expects.

And all of this — for every lesson, on the same day, by the same rules, without fatigue and without "favorites".

The result does not need to be exported manually anywhere — it appears in the web panel (admin panel). The owner, the academic manager, and the teacher each log in under their own role and see exactly what concerns them.

   Lesson recording (video + audio from the camera)
              │
              ▼
        ControlAI analyzes it
              │
              ▼
   Ready report in the admin panel  →  owner · academic manager · teacher

Exactly how ControlAI "listens to" and "watches" a lesson (the technologies behind it) — we deliberately do not cover that in this document: here we explain what the product does and why, not how it works on the inside.

Before → after¶

The best way to understand ControlAI is through what it replaces.

In many centers, lesson quality is controlled by a live person — the "nazoratchi" (observer, methodist). They sit down at a monitor, selectively watch lessons over the camera, manually fill in an observation form, and draw conclusions.

	Before: manual control	After: ControlAI
Coverage	1–2 lessons out of a hundred — there is no time for more	Every lesson, no exceptions
Speed	Hours of manual viewing	Report ready about an hour after the lesson
Consistent rules	Subjective, depends on the person and their mood	The same criteria for all lessons
What is measured	By eye, "liked it overall"	Precise numbers: talk-time, questions, attendance, punctuality
Scale	The bigger the center, the less you can keep up with	10 classrooms or 100 — the workload is the same

Important: the human does not disappear. The academic manager is still needed — but their role changes. Previously they spent hours watching lessons from scratch in order to fill in the form. Now ControlAI analyzes all the lessons and fills in the form itself, and the academic manager is left to review the finished result and correct the questionable parts. And yes — if the center has enabled mandatory review before publishing (publishing can also be entrusted to the system — that is a per-center setting), they still go through lessons "one by one", but this is no longer an hour-long breakdown of the recording, it is a quick review of the finished result: the AI assigns most of the scores confidently, those just need a quick confirmation, and attention goes only to the doubtful items. This is why the academic manager manages to cover all the lessons, not just a couple, and does expert work rather than routine.

Three sides of one lesson¶

ControlAI looks at a lesson from three sides at once — and together they form the full picture. Some signals already work in the pilot, some are on the roadmap (marked separately), but the idea is the same: extract the maximum amount of useful information from an ordinary lesson recording.

1. Speech — what happened during the lesson¶

who spoke and how much — the teacher or the students (talk-time balance);
how many questions the teacher asked;
how many pauses and how much silence there were — whether the lesson "sagged";
target-language ratio — for example, how much of the lesson was in English (or another target language) and how much in the native language;
punctuality — whether the lesson started and ended on time;
(roadmap) rude or inappropriate language — whether anyone swore during the lesson; praise and encouragement of students; addressing students by name; distractions onto off-topic subjects.

2. Video — what happened in the classroom¶

how many students were present (attendance);
how many students were late and by how much;
empty classroom — if the lesson did not actually take place, the system sees it;
(roadmap) who is distracted by their phone; who participates actively and who is passive the entire lesson; head on the desk (signs of sleep/fatigue); how often students raise their hands.

3. Methodology — whether the lesson met the center's standards¶

ControlAI fills in the center's own rubric (the lesson observation form);
checks whether the teacher covered the planned topic and used the accepted methodology;
tracks topic coverage — what has already been covered and what has not yet.

The full list of what ControlAI can do is in the chapter "What it can do"; what exactly each number means is in the chapter "Metrics and what they mean"; and what does not work yet and will appear later is in the chapter "Roadmap".

Optional: teacher voice ID (Voice ID)¶

So that the system can more accurately distinguish where the teacher is speaking and where the students are, the teacher can record a sample of their voice once. Right in the admin panel, in the personal profile, there is a record button for this: you need to tap the microphone and read a short text — in several variations (normal tone, louder, slow, fast) and in different languages (Uzbek, Russian, English). The more varied the samples, the more accurately the system recognizes the voice.

This is optional — the teacher decides for themselves whether to record a sample or not. Right now only the teacher records a voice sample; voice recognition for students is on the roadmap. When it appears, personal reports for each student will become possible (who spoke how much, how many times they answered) as well as reports to parents. More on this in the chapter "Roadmap".

What ControlAI does NOT require from the center¶

No new equipment is needed in the ordinary case — the same cameras that already record the classrooms are used.
Nothing needs to be installed for the students — no apps, badges, or sensors.
No reviewer presence at the lesson is needed — the teacher runs the lesson as usual.

ControlAI fits into what already works at the center and starts delivering value without restructuring the teaching process.

Since ControlAI works with the lesson recording (including the speech of the teacher and the students), launching it requires consents from the teachers and — where required — the parents. How this is arranged and how the data is protected is covered in the chapter "Privacy and trust".

The key takeaway from this chapter: ControlAI automatically analyzes every lesson from the recording made by the already-existing cameras and shows the result in the admin panel — replacing slow and selective manual control with fast and comprehensive control, while not removing the live academic manager but freeing them from the routine.

→ Next: 02. Why ControlAI is needed — which pain points of the centers it addresses and why "just dumping the video into any AI" is not enough.