See who’s
actually in the room.

When class goes online, the room goes dark.

In person, an instructor reads the room without thinking, faces, posture, who’s leaning in, who’s lost. In a large live online class, those cues shrink to thumbnails or vanish behind black tiles. Everyone looks equally present, whether they’re locked in or long gone. And learners feel it: they tell us they still want to finish the course and still prefer the pull of in-person learning, but online they feel unseen when no one can tell they’re confused, passive, or falling behind.

Disengagement is real, and it’s silent.

Across our interviews, the same pattern kept surfacing: learners still value finishing the course and still prefer the engagement of in-person learning, yet online they drift in ways no instructor can see. They feel unseen when no one can tell they’re confused, passive, or falling behind, and the one instructor we spoke to felt exactly that blind spot from the other side.

One room. Two people in it.

We deliberately narrowed from “all online learning” to a single, well-defined relationship: a student and an instructor in the same large, live online class. It’s the one stakeholder pair we have two-sided evidence for, and the one we can actually build and demo.

Sense. Surface. Respond.

Our working hypothesis: a lightweight layer on the video call that reads a few clear attention signals, shows the instructor the room, and hands them the nudge, without ever taking the teaching decision away. This is the direction we’d build and test first, not a finished spec.

For the first version we deliberately read a focused set of signals. Each one only flags after it holds for 30-45 seconds, so a glance at your notes or a stretch never trips it.

These surface how present someone looks, not what they’re thinking or how they’ll be graded. Every flag is a probabilistic attention signal for the instructor to interpret and act on, never a penalty or an automated judgment. How we detect them reliably, an off-the-shelf model or a simpler rules-based proxy for the demo, is exactly what we’d settle first.

Two views, one signal.

The student attends class as normal and only ever gets a quiet, private nudge. The instructor sees the room in real time, each student’s engagement state, a live class score, and a copilot that steps in when attention slips. This is the real working product, running live right here, click either view to open it.

Built opt-in, not surveillance.

Reading attention from a camera is sensitive, and we treat it that way. Privacy isn’t a footnote here; it’s a design constraint we’d hold from day one. If we can’t do this without feeling like monitoring, it shouldn’t ship.

Everyone reports after class. No one shows the room during it.

The engagement tools we benchmarked are dashboards you read after the session, completion rates and post-hoc analytics, mostly aimed at learners. Our bet: the moment that matters is during class, in the instructor’s hands, while they can still change the next ten minutes. That’s the gap we’d aim for, and the assumption we most want to pressure-test.

We’re holding the problem tightly, the solution loosely.

The problem is real and clear. How we solve it is still a set of bets, and these are the ones we most want to pressure-test before committing. Tell us which we’ve got wrong.

Start free. Scale when it clicks.

Directional, not final. Once the diagnostic genuinely works, this is one way the model could pay for itself, a free pilot for a single class, paid depth for a teaching team, and a campus-wide tier for institutions, who are the real buyer here.

The whole AHA story, one click away.

The pitch, the story walkthrough, the live product, and the strategy behind it, roadmap, metrics, and go-to-market.