The recovery layer your LMS doesn’t have.
- Student success & tutoringAdvisors, SI leaders and tutors decide who to call.
- Zelvi — the recovery layerMisconception → re-teach → retry → recorded result → hand-off with context.
- AI assistantsExplain, summarise and generate practice.
- Course materialYour syllabus, OER, publisher content, SOPs.
- LMSCourse shell, roster and gradebook.
The moat
Four things that compound with every course.
Grounded, or it declines.
Ava answers only from the material the instructor published and cites the page. When the material does not cover it, the student is routed to the instructor instead of a guess.
Answers that stayed inside the material, on our go/no-go benchmark.
A recovery record, not a transcript.
Every struggle ends in a retry and one of three results, per misconception. Silence is never counted as success, so the numbers a dean sees cannot flatter themselves.
Rule first, AI second.
Answers are checked by rule, then a hint from your material, and only then does the model explain. Cost scales with struggle, not with how much students chat.
- AttemptNo AI
- Hint from your materialNo AI
- Ava explainsAI call
The hand-off carries the evidence.
A student still stuck reaches the instructor or a course assistant with every attempt attached, and faculty see what to reteach, by concept, while the term is running.
Ana C. asked for help · Slope from two points
2 attempts · misconception: subtracted in opposite orders · Ava’s explanation attached
The loop
What happens after the wrong answer.
- 1Spots the struggleA wrong answer in practice, or a student telling the tutor they are lost at midnight. Answers are checked by rule first, so the easy cases never need the AI.
- 2Names the misconceptionThe wrong answer is matched to a meaning your faculty signed off, so "wrong" becomes "sign error on the slope".
- 3Re-teaches it from your materialAfter two missed tries, Ava explains it by voice or text and draws it on the whiteboard, citing the page in your course. If your material does not cover it, she says so.
- 4Makes the student try againAnother attempt on the same idea, by the student. An explanation is not evidence of learning; a fresh attempt is.
- 5Records what happenedRecovered, repeated the same mistake, or not seen since. Silence is never counted as success.
The platform
Five parts, one loop.
- AvaA tutor that knows which course it is teaching.Ava answers from what your institution published, cites the page, and declines rather than improvising when the material runs out.Read more
- Visual CanvasShow the idea. Then let them move it.Graphs, circuits, reactions and anatomy the learner can drag, label and re-run — generated by the product, not pasted in as a picture.Read more
- Curriculum GroundingThe decline is the product.A general assistant answers everything, so it can never tell you what your material fails to cover. Ours can, and it tells the instructor too.Read more
- Practice & AssessmentThe model observes. The rubric decides.A competency result that cannot name the rubric version it was judged against is not evidence. Ours can.Read more
- Live GamesThe game ends. The learning doesn’t.Students join on their phones with a code or a QR. When the last question closes, every miss becomes that student’s own practice, with a hint first and the explanation after.Read more
- AnalyticsA zero can mean two opposite things."Nobody has practised yet" and "they practised and struggled" look identical in most dashboards and call for opposite responses. We keep them apart.Read more
A glimpse of what Ava draws, live
Concepts they can take apart.
Every figure here is rendered by the same code path that runs for a real student. Not screenshots. Not stock illustrations. Drag one, and it moves.
The canvas, close up
Every figure is built from the concept.
A diagram about the exact function, circuit or pathway your document covers, instead of the nearest stock picture. Drag a point and the rest moves with it.