Skip to content

What ControlAI is

In short

ControlAI is a class-session analytics system for universities.

Cameras are already installed in most classrooms. But almost nobody watches the recordings — they're a "dead archive" people return to only after something has already gone wrong. ControlAI takes those recordings and turns every class session into a clear report: did the session actually happen, did it start and end on time, who talked and how much, how many questions were asked — and did the class meet the university's methodology requirements. Over time it will also learn to count student attendance and late arrivals from video, and to notice phone distractions (this is on the roadmap).

Normally no new equipment is needed. ControlAI connects to the cameras and video recorder already running at the university, and reviews every single class session — not just once or twice a semester, the way an open (demonstration) lesson or peer observation does.

A bit more detail

Imagine you had a very attentive assistant that could:

  • "listen" to a class session and tell who spoke and how much (the instructor or the students), how many questions and pauses there were, and what language was used (Uzbek, Russian, or mixed) — and over time (roadmap) also pick up on praise or inappropriate language;
  • confirm the class actually took place — whether it fell through, whether it started on time, or whether students were let out well before the end (a standard class session is 80 minutes, and the recording shows how much of that time was actual teaching);
  • check the session against the university's methodology requirements — whether the instructor covered the planned topic the way the department expects.

And all of this — for every class session, on the same day, by the same rules, without fatigue and without playing favorites.

The result doesn't need to be exported anywhere manually — it shows up in the web panel (admin panel). The quality-control office, deans' offices, heads of department, and instructors work in the panel every day and see exactly what concerns them; the vice-rector for academic affairs pulls a weekly summary from it, and the rectorate gets the finished results. For leadership this is objective data; for an instructor it's personal statistics and an automatic summary of every class session: a development tool, not just a control mechanism.

   Class-session recording (video + audio from the camera)
              │
              ▼
        ControlAI analyzes it
              │
              ▼
   Finished report in the admin panel  →  quality office · dean · head of department · instructor · rectorate

Exactly how ControlAI "listens to" and "watches" a class session (the technology behind it) is deliberately not covered in this document: here we explain what the product does and why, not how it's built under the hood.

Before → After

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

At universities today, class quality is monitored manually and selectively: open (demonstration) lessons once or twice a year, scheduled peer observations, and checks by the quality-control office. A reviewer sits in on the class, fills out a lesson observation form by hand, and draws conclusions — for one class session out of the hundreds the university runs every day.

Before: manual review After: ControlAI
Coverage 1–2 class sessions per instructor a year — there's neither the people nor the time for more Every class session, no exceptions
Speed Sitting in on the class plus hours writing it up The report is usually ready within an hour of the class (by morning if the connection is weak)
Consistent rules Subjective, depends on the reviewer and their mood The same criteria for every class session
What's measured By eye, "overall it was fine" Exact numbers: whether the class happened, timing, talk time, questions
Observer effect The instructor prepares a "showcase" lesson for the committee's visit An ordinary session on an ordinary day — the real picture
Scale The more faculties, the less you keep up 10 classrooms or 100 — the same workload

Important: the human doesn't disappear. The quality-control office and heads of department are still needed — but their role changes. Before, the reviewer had to sit through the class session from start to finish just to fill out the form. Now ControlAI reviews every class session and fills out the form itself, leaving the expert to review what's already done and correct anything questionable. If the university has turned on mandatory review before publishing (auto-publish is also an option — it's a setting), the expert just quickly confirms the AI's confident assessments and looks closely at the uncertain ones. That's why the quality-control office manages to cover every class session instead of a handful a year, and spends its time on expert judgment rather than routine work.

Three sides of one class session

ControlAI looks at a class session from three angles at once — and together they form the full picture. Some signals already work today, others are on the roadmap (marked separately), but the idea is the same: squeeze the most useful information out of an ordinary class recording.

1. Speech — what happened during the class

  • who spoke and how much — the instructor or the students (talk-time balance: predictably one-sided in a lecture, quite different in a seminar or a practical class);
  • how many questions the instructor asked;
  • how much pausing and silence — whether the session dragged;
  • the language of the class — which language was spoken (Uzbek, Russian, or a mix; for language courses, the share of the target language);
  • punctuality — whether the class session started and ended on time, and whether the group was let go well before the end;
  • (roadmap) inappropriate or offensive language — whether any was used during the class; praising and encouraging students; addressing students by name; drifting off into unrelated topics.

2. Video — what happened in the classroom

  • an empty classroom — if the class session didn't actually happen, the system sees it and flags it;
  • (phase 2, roadmap) attendance (davomat) — how many students were present and how many arrived late; who is distracted by their phone; who is actively participating versus passive the whole class; head down on the desk (signs of sleepiness/fatigue).

3. Methodology — did the class meet the university's standards

  • ControlAI fills out the university's own lesson observation form — essentially an open (demonstration) lesson, except not once or twice a year but for every class session;
  • checks whether the instructor covered the planned syllabus topic using the accepted methodology;
  • tracks topic coverage — what has already been covered in the course and what hasn't.

The full list of what ControlAI can do is in the "What it can do" chapter; what each number actually means is in "Metrics and what they mean"; and what doesn't work yet but is coming later is in the "Roadmap" chapter.

Optional: instructor voice sample (Voice ID)

To let the system tell the instructor's voice apart from the students' more precisely, the instructor can record a voice sample once. There's a recording button right in the admin panel, on the personal profile page: tap the microphone and read a short passage — in a few 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 instructor decides for themselves whether to record a sample. Only the instructor records a voice sample: the system does not identify individual students by voice — the subject of the analysis is always the class session and the instructor's work, never specific students (see the "Boundaries" chapter).

What ControlAI does NOT require from the university

  • No new equipment is needed in the typical case — the same cameras and video recorder that already record the classrooms are used. The one possible addition is an external microphone plugged into the camera's audio input where the built-in mic fails the audio test (say, in a large lecture hall); the audio audit done at connection time decides this.
  • Nothing needs to be installed for students — no apps, badges, or sensors.
  • No reviewer needs to be present at the class session — the instructor teaches as usual, with no "showcase" sessions for a committee.

ControlAI plugs into what already works at the university and starts delivering value without restructuring the academic process.

Because ControlAI works with the class recording (including the speech of both the instructor and the students), launching it requires instructor consent and student notification — since students are adults, the consent process is simpler. How this works and how the data is protected is covered in the "Privacy and trust" chapter.

The key takeaway from this chapter: ControlAI automatically reviews every class session from the recording of cameras already in place and shows the result in the admin panel — replacing rare, selective manual review (open lessons, peer observation) with something fast and comprehensive, without removing the human experts — the quality-control office and heads of department — but freeing them from routine work.

→ Next: 02. Why a university needs ControlAI — what pain point it solves for universities, and why "just dump the video into any AI" isn't enough.