tutor
One question
Only the student who opened chat. Tomorrow’s class stays the same.
VLearn
From an idea to the real classroom
meet → open the tools
Quiet class
Hand signals
Five questions
One day
meet · the problemVinUni
beyond a chatbot
Cognitive
silence.
one student asks
End-of-term surveyWeek 15. Too late for this cohort.
Students don’t raise a handThe class, and Teams, stay quiet.
Chat only hears who spokeSilence leaves no question.
Open commentsFree text piles up. Read by hand, weeks later.
meet · why one productfully adaptive
tutor · quiz · KC · tracking — each piece alone
Close the loop.
quiz · KC
Knows the gap
Not the page, or the second, where the class got stuck.
tracking
Has logs
Not yet the thing an instructor does the next morning.
one product
Adaptive enough
Signal → where the class is stuck → tutor and quiz on that KC → a person changes the lesson.
HCI · signals from the hands, not the mouth
dwell
page back
selection
lasso
ask AI
help request
codelab
anonymous feedback
Dwell is a weak signal. Asking and feedback are stronger. The heatmap is the cohort, not one named student.
meet · the projectone day, many questions
not one person’s feature
Learning day
Author a day together. Slides, video, lab rooms, quiz. More than one instructor can be in the draft.
Learner model
Five questions about one student. Not a single score, and not a copy of one paper.
Profile
Evidence, a scope, and an expiry. A wish is not an observation. The student can contest it.
Concepts
Drawn from the chapter itself. Semantic match can be high. Matching an expert index stalls near 50%.
Tutor
Help on the slide and the video second. It proposes a move. It does not write mastery.
meet · learner modelfive questions
what an instructor can ask about one student
Coverage
What they reached and finished. Video, document, lab, and quiz stay separate. Unmeasured stays unknown.
Process
The order of the work. Start, answer, hint, finish, retry, resubmit. A timeline, not a total.
Support
Where help was needed, and whether it arrived. No data means unknown, not “fine”.
Mastery
An estimate per student and KC, with history, source, and model version.
Evidence confidence
How many tasks, how fresh, what is missing. No single validated score yet.
working session · researchwhat was actually found
drafts and limits, not a finished learner
Concepts from the chapter
Semantic precision about 83–90%. After a cheap filter, purity 94–100%. Expert-index F1 stalls near 52–55%, and that ceiling is structural.
Profile is four different things
A wish, an observation, an estimate, and the next action. “Dislikes Socratic” may mean “in a hurry”.
Affect is a proposal
Confusion, frustration, uncertainty, from the dialogue only. No camera. It must not rewrite mastery, and it is not running on students.
Signals are candidates
Time on the material, coverage, reread, skip, time-to-first-help, codelab runs and test failures. Not one formula. Dwell alone is weak.
working session · where it standsclassroom and research
in the classroom
The day
is real.
Authoring
Shared draft. Presence shows who else is editing.
Lab
Rooms from the database. Raise hand. A note on every bonus.
Reader
The slide on screen, the video second, text from a region.
Signals
Consent at login. No silent grant. Cohort, not a name.
still research
Not a finished
learner.
Five states
Local audit and replay. Not a full lesson. Not production.
Profile
A design to discuss. More clicks are not an adaptive student.
Concepts
Real concepts hold. Index match stays near 50%.
Affect
Three labels, dialogue only. Must not edit mastery.
working session · demoalt-tab
Let's open the hood
Dive into the actual tools & workflow
← → · screen edges