See what your classroom is really feeling.
Vignan ClassPulse AI is an agentic vision system that watches a lecture recording and tells teachers — frame by frame — exactly when students were engaged, distracted, drowsy, stressed or on their phones.
No identities stored · anonymous per-student tracking · faculty-only access
Faculty lack real-time visibility into classroom engagement
Teachers can only watch a few rows at a time. Yet engagement, attention, and emotion drive every learning outcome.
No visibility
A 60-student classroom is impossible for one human to read in real time.
No evidence
Faculty rely on intuition. Curriculum changes happen without behavioural data.
No feedback loop
By the time exam scores arrive, the lecture is months gone.
An agentic AI that watches every frame for you
Drop yesterday's recording. Get a detailed narrative report with peak/low engagement windows, attendance-vs-engagement correlation and concrete teaching recommendations.
Upload the recording
MP4 from any phone, webcam or lecture-capture system. Frames are extracted client-side — no raw video leaves your browser.
Vision agents analyze
Advanced vision AI scores attention, emotion, stress, drowsiness, phone use, group interaction and instructor mobility — anonymously, every 30s.
Get an actionable report
Narrative summary, peak / low engagement windows, attendance correlation and AI-generated teaching recommendations.
9+ behavioral signals — every signal a great teacher wishes they had
Happiness index
Per-frame positive affect across all visible students.
Attention & engagement
Whether students look at the instructor, screen or task.
Stress & drowsiness
Withdrawn, anxious or fatigued behavioral signals.
Phone usage
Counts students visibly using a phone in each frame.
Group interaction
Collaborative behavior and peer-to-peer communication.
Per-student tracking
Anonymous S1, S2, … tags — no identities stored.
Attendance correlation
Students-visible vs engagement timeline overlay.
Instructor mobility
Detects whether the instructor moves around or stays static.
Note-taking signal
Estimates active note-taking density per frame.
Production-grade agentic architecture
Multi-agent vision pipeline, edge-deployed inference, real-time streaming reports.
1 · Client-side frame extractor
- Browser-side ffmpeg / canvas pipeline
- Samples 1 frame every 30s
- Raw video never leaves the device
2 · Vision agent
- Per-frame structured JSON scoring
- 9+ behavioral signals + per-student tags
- Streaming results into the cloud database
3 · Cloud database (Postgres + RLS)
- frame_analyses + video_reports tables
- Row-level security · faculty-only access
- Realtime subscriptions for live dashboard
4 · Finalize-report agent
- Aggregates frames into narrative
- Generates peak/low windows + recommendations
- Edge function — sub-second cold start
Enhanced teaching effectiveness — measurable, repeatable
Faculty get behavioural feedback in minutes, not semesters.
From happiness index to phone usage — every dimension the brief asks for, and more.
Anonymous S1, S2… tags. GDPR-friendly. Faculty-only RLS access.
What teachers ask first
No identities are stored. Students are tagged anonymously as S1, S2… per video. Only aggregate behavioural scores persist.
Yes — frames stream in and dashboard charts update in realtime via Supabase realtime subscriptions.
Any MP4 from a phone, webcam or lecture-capture system. Frames are extracted in your browser — raw video never leaves the device.
Only the faculty owner. Row-level security enforces access at the database layer.