The path of one class session¶

Here is the whole journey of a class session: from the moment it begins to the finished report in the panel. We'll first walk through it with a live example, then show the same thing as a diagram, and then lay it out step by step. No technical detail — just what happens, and in what order.

The story of one class session¶

Names and numbers are invented, for illustration.

Monday, 2:00 PM. In classroom No. 305, a 2nd-year group (the "Logistics" program, Russian-language track) starts a practical class in "Transport Operations" — a standard 80-minute class session. It's taught by instructor Karimova D. (assoc. prof.). As usual, an ordinary camera hangs in the classroom — the same kind already installed in typical universities — recording the session (video and audio). Neither Karimova nor the students do anything special: the session runs as it always does.

3:20 PM. The session ends, students leave. This is where human involvement in the "magic" ends — from here on, ControlAI does everything.

3:21 PM. From the class schedule, the system knows that the 2nd-year group's session in classroom No. 305 has just ended. It retrieves the recording of the session itself, from the university's video recorder (NVR/DVR) — the same one the cameras record to anyway; nobody uploads files by hand.

Next, ControlAI analyzes the session from several angles at once:

it listens — recognizes the speech and works out who spoke (Karimova or the students) and for how long, how many questions there were, and how many pauses; it also reliably handles mixed Uzbek-Russian speech;
it checks against the observation form — fills out the university's own lesson observation form: was the planned topic covered the way the department expects;
it watches (stage 2) — at the next stage of the system's development, the same video will add attendance (davomat) and student late arrivals; the recording for this is already being captured today.

4:20 PM, roughly an hour after the session. The report is ready and shows up in the panel. Everyone sees what's relevant to them:

The dean and the dean's office see the lesson card: the session took place, started and ended on time, the instructor spoke 58% of the time, students — 42%, 24 questions were asked. Alongside it — the "Lesson DNA" (a colored ribbon showing how the time was split) and a few flagged moments.
The quality-control office (ichki ta'lim sifati nazorati bo'limi) opens the completed observation form, quickly reviews the AI's assessments, and corrects them if needed.
The head of department sees this session within the overall picture of their department — alongside the rest of the sessions and instructors.
Karimova logs into her personal account and sees her own statistics and a short summary of the session — not a "grade handed down from above," but data that helps her understand what to improve.

Nobody rewatched the recording. Nobody sat through an "open lesson" with a form in hand. The session simply turned itself into clear numbers and a report — the same way, for every class session in every classroom.

The same journey — as a diagram¶

flowchart TD
    A["Class session is in progress — the existing camera records video and audio"] --> B["Class session ended (per schedule)"]
    B --> C["ControlAI retrieves the recording from the university's video recorder itself"]
    C --> D["Session analysis"]
    D --> D1["Listens: who spoke, questions, pauses"]
    D --> D3["Checks against the lesson observation form"]
    D -.-> D2["Stage 2: attendance and late arrivals from video"]
    D1 --> E["Finished report — in about an hour"]
    D3 --> E
    D2 -.-> E2["Video metrics — will supplement the report at stage 2"]
    E --> F["Panel"]
    E2 -.-> F
    F --> G1["Dean's office / dean"]
    F --> G2["Quality office / head of department"]
    F --> G3["Instructor"]

The same journey — step by step¶

The class session is in progress. The camera in the classroom records video and audio — as usual, with no human involvement. It uses the cameras and video recorder already present at the university.
The class session ends. From the class schedule, ControlAI knows when and in which classroom a session has ended — the schedule works as the system's "control panel."
The recording is retrieved. The system downloads the recording of the session from the university's video recorder itself — nothing needs to be uploaded manually.
Audio becomes speech. ControlAI "listens" to the session: it recognizes what was said (including mixed Uzbek-Russian speech) and separates who spoke — the instructor or the students.
Video becomes the classroom (stage 2). At the next stage, the system will learn to "watch" the footage — separately, and without delaying the report: how many students are in the classroom, how many arrived late. The recording for this is already being collected today.
The metrics are calculated. Whether the session actually took place, started and ended on time; talk-time balance, number of questions, pauses, punctuality — all expressed in numbers.
Checked against the observation form. Fills out the university's lesson observation form and checks whether the planned topic was covered.
The report is written. Turns the finished numbers into a short, clear text — in the group's language (Uzbek or Russian).
Shown in the panel. The result appears in the panel — everyone sees, under their own account, what's relevant to them.

What if the class session didn't happen?¶

There are two cases here.

The session was canceled in advance. If it's already known that a session won't take place (a holiday, a substitute instructor, a stream reschedule), this is marked in the schedule in the panel ahead of time — and ControlAI simply doesn't take that session into processing at all.

The session simply didn't happen. If the session is still on the schedule but no one showed up or the classroom is empty, ControlAI notices this on its own: there's no speech in the recording, so it doesn't spend time on a full analysis and marks the session as "not held," and that session isn't included in the billed volume. The dean and the dean's office see the mark and understand what happened: this is the objective answer to "did the session actually take place?" — for every session, with no inspectors in the hallways.

Lectures and seminars¶

Monitoring whether a session was held, its timing, and the lecturer's share of speaking time — all of this works in any classroom, from a small seminar room to a large lecture hall. A fair caveat: transcript completeness in large lecture halls depends on how well the camera "hears" the room. That's why an audio audit is carried out for every classroom during onboarding: wherever the camera's built-in microphone fails the test, an external microphone is attached to the camera's audio input — that's the only thing that might be added to a classroom.

How long the results take¶

The main report for a class session (speech, questions, the observation form) is usually ready within an hour of the session ending. Video metrics (attendance, late arrivals) are stage 2: once they're available, they'll arrive later than the speech metrics, within a day, supplementing the report and never delaying it.

If a building's network connection is weak, the recording may not be retrieved right away but overnight instead — in that case the report will be ready the next day. This is agreed on during onboarding.

Key takeaway from this chapter: once a class session ends, ControlAI retrieves the recording itself from the video recorder already present at the university, listens to the session, calculates the metrics, checks it against the observation form, and usually shows the finished report in the panel within an hour — separately for the rectorate, the dean, the quality office, the head of department, and the instructor (video-based attendance will be added at stage 2). Nothing is installed in the classrooms, and no one rewatches recordings or fills out forms by hand.

→ Next: 04. What ControlAI can do — the full list of what the system sees and measures.