Roadmap¶

This chapter gathers everything the document flagged as "roadmap" — what's coming later. We think of ControlAI as building a university's smart academic-process system ("Smart o'quv tizimi") stage by stage: the first stage is working today, and later stages grow the system on top of the same infrastructure. This is a direction of development, not a promise of specific dates — the order and scope may change.

Stage 1 — available now¶

Everything listed in the previous chapters isn't a plan — it's the working first stage:

speech analytics for every class session: transcript, instructor and student talk time, questions, pauses;
class-session conduct monitoring: lesson not held, the instructor was late, dismissed the group early — objective timing pulled from the recording, with no observer in the room;
methodological review (an open lesson on every class session): the system fills out the university's own lesson observation form for every class session — not once or twice a year by a committee;
instructor rating 1–10 and comparative analytics: department vs. department, faculty vs. faculty, semester dynamics;
a panel with roles for the rectorate, deans, departments, and the quality-control office; and personal statistics for the instructor.

The university doesn't start with a "pilot of future technology" — it starts with a finished, working stage, and already at this stage it accumulates objective data on teaching quality.

Stage 2 — attendance and video metrics¶

The cameras in classrooms record more than just audio, and the next stage is video analytics from those same cameras — no new equipment:

attendance (davomat): how many students are in the room (counted from video frames), student latecomers, an empty classroom;
attendance dynamics by group, course, and faculty.

On top of attendance — separate validated detectors:

how many students are distracted by their phone;
how many students are active, and how many are passive for the whole class session;
head on the desk (signs of sleeping or fatigue);
how often students raise their hand (in seminars and practicals).

To be clear: today this is not a current feature — it's the next stage. Speech reports and class-session conduct monitoring don't wait for it.

Stage 3 — HEMIS integration and rectorate dashboards¶

HEMIS integration: the class schedule and student enrollment are pulled in automatically from the government system, instead of manual entry in the panel; the system picks up schedule changes (a rescheduled class, a substitute instructor) on its own, so analytics always relies on up-to-date enrollment;
consolidated rectorate dashboards (a basic university-wide summary already exists today — this stage adds a single screen and trends): teaching quality and class-session conduct across the whole university — faculties, departments, semester dynamics; the vice-rector sees in a minute where everything is fine and where it's worth a closer look, and brings a ready-made summary to the rectorate meeting.

Until this stage, the schedule is entered in the panel — including via import from Excel/Google Sheets, such as an export file from HEMIS, which means less manual work during onboarding.

Finer-grained speech analytics¶

language share throughout the class session (how much Uzbek, how much Russian, how much English — by segment);
interaction pace (how many "instructor ↔ student" exchanges per 10 minutes);
how talk time is distributed among students ("3 students spoke 90% of the time" — important for seminars);
audio-quality monitoring in the classroom (catches a "dying" microphone early);
signals from speech: rough language, praising students, addressing students by name, drifting off-topic.

Deeper on methodology¶

calibrating metrics by class type: in a lecture, a long monologue from the instructor is normal; in a seminar, the same talk-time balance is worth a closer look. Norms for lectures, seminars, practicals, and labs will be calculated separately, making instructor comparisons more accurate;
lesson phases (introduction → delivering material → practice → independent work) as a separate signal and a timing check;
coaching tips for the instructor, generated from the observation form;
weekly summaries with trends, trend alerts (the rating has been dropping for several weeks in a row), comparison against the department and university average;
semester-end reports: semester results by instructor, department, and faculty — and comparing semesters against each other: whether teaching quality improved compared to the previous semester, and exactly where.

Convenience and accuracy¶

continuously improving recognition of Uzbek-Russian speech, based on corrections and labeled data that accumulate in the system (the recordings themselves keep being deleted on the schedule set out in the Privacy and trust chapter); the longer the system runs, the more accurate it gets — and the more valuable your university's own accumulated history becomes.

What we deliberately do NOT do¶

We've deliberately postponed or excluded some things — for reasons of accuracy, cost, or ethics:

an open "ask anything about the video" feature (query any moment) — unreliable, expensive, and unacceptable from a privacy standpoint;
assessing emotion/engagement from faces — ethically contentious;
automatically grading students and exam proctoring — the system analyzes the instructor's work during the class session, it doesn't examine students;
a public instructor "leaderboard" — it's demotivating and contradicts the "helper and transparency, not surveillance" approach;
"live" analysis — the product reviews a class session that already happened, it doesn't stream one.

The main point of this chapter: the smart academic-process system is built stage by stage: stage 1 (already working) — speech analytics, class-session conduct monitoring, methodological review of every class session, ratings, and a panel for the rectorate, deans, and departments; stage 2 — attendance and video metrics; stage 3 — HEMIS integration and consolidated rectorate dashboards. Beyond that — finer-grained speech and methodological analytics: calibrating norms by class type, semester-end reports, growing recognition accuracy. And a number of things we deliberately don't do — for reasons of reliability, cost, and ethics.

→ Next: 17. Glossary — a short glossary of terms.