AffectLearn senses the moment you get stuck or start drifting, then reshapes the lesson in real time. A hint when you're lost, a nudge onward when you've stalled, a breather when you need one. All from a standard webcam and your browser.
Built for self-paced learners and the educators who design their courses
Every 30 seconds, AffectLearn reads subtle signals — facial expression and how you move through the page — to understand your state and gently adjust what you see next.
Reads facial micro-expressions and your reading rhythm — scroll pace, dwell time, hesitation.
Each signal answers the question it is good at — your camera reads drifting attention, your typing and scrolling read confusion.
Weighs your state against your progress and pace to choose the most helpful next move.
Offers a hint, a different explanation, a move onward, or a well-timed break — always as a suggestion you can wave away.
Not before, not after. AffectLearn responds to your real state in the moment so momentum never breaks.
Re-reading the same paragraph? A focused hint or simpler analogy slides in to unblock you, then steps aside.
Breezing through familiar material? The lesson skips the obvious and offers a harder, more rewarding problem.
When frustration builds, AffectLearn suggests a short reset, and returns with a fresh way to explain the idea.
If dense text isn't landing, the same concept reappears as a visual walkthrough or a hands-on exercise.
Every adaptation is a suggestion you can accept or dismiss. AffectLearn guides, it never takes the wheel.
Prefer camera off? The platform reads your interaction patterns instead and keeps adapting, with full transparency.
A small, always-visible panel shows your current state and that detection is running on-device — never a black box.
When confusion lingers on a tough concept, a gentle hint appears beside the content — not a pop-up that interrupts.
Attention drifting for more than a moment nudges the lesson forward rather than leaving you stuck on a page you have stopped reading.
Study independently without feeling alone. AffectLearn notices when you're drifting and meets you with the right support — so you keep going and actually finish.
Not just whether they passed — but which section lost them, and how. Turn real emotional signals into precise, confident course improvements.
Reading your expressions doesn't mean recording your face. Frames are analyzed the instant they're captured and immediately discarded — nothing is saved, sent, or seen by a human.
Webcam frames never touch disk or database. Processed in memory, then gone.
Turn the camera off anytime. The platform keeps working in behavioral-only mode.
Your learning signals stay private — never sold, never linked to advertising profiles.
Everything in transit and at rest is encrypted to modern standards. Clear consent, always.
Affect detection is easy to claim and hard to do. Here is what ours actually scores, on public benchmarks, with the sample behind each number.
Measured on EngageNet across 26 held-out participants — people the model had never seen. The best published result on the same benchmark is 0.74. The model reads facial geometry, not pixels: about 600 bytes leave your browser per cycle, against the ~3.3 MB a face crop would cost.
Measured on DUX across 46 annotated sessions, from mouse, scroll and typing rhythm alone. On the one dataset that carries both signals, interaction beat facial expression for confusion — which is the opposite of what most of the field assumes, so we publish it rather than bury it.
Whether adaptation improves learning outcomes is an open question here. The trial that would answer it has not run, so we do not claim a result. Two of the four classic states — boredom and frustration — have no detector we would stand behind either, so the system acts on the two it can actually see.
No. Frames are analyzed in real time, the moment they're captured, and immediately discarded. No video is ever written to disk, transmitted, or reviewed by a person. Only an anonymous read of your current state is used to adapt the lesson.
That's completely fine. AffectLearn switches to behavioral-only mode and reads how you move through the lesson — reading pace, hesitation, navigation patterns — to keep adapting. You'll always see clearly which mode you're in.
Two, and we would rather say so than claim four. Confusion is read from how you work — mouse, scroll and typing rhythm — and drifting attention is read from the camera. Frustration and boredom have no detector we would trust, so the system does not pretend to see them. Confusion brings a hint; drifting attention nudges the lesson on.
Never. Every adaptation is a suggestion you can accept or dismiss. Hints appear calmly beside your content rather than as blocking pop-ups, and you stay in control of your path through the material.
Just a modern browser and, optionally, a standard webcam — nothing to install. After a quick 30-second calibration where you read a short paragraph, the platform establishes your baseline and you're ready to learn.
Structure your material as lessons, sections, exercises and quizzes, then upload it through the creator dashboard. As learners progress, you'll get per-section affect analytics showing exactly where they struggle — so you can improve with evidence.