Problem
Taking attendance by hand wastes class time and produces records that are easy to lose, miscount, or backfill. In larger classes the cost compounds — minutes lost every session, and no reliable signal about who is actually showing up over a semester.
CObot+ set out to remove the manual roll-call entirely while keeping the records trustworthy.
Approach
The system pairs an on-site capture device with a cloud backend. The edge device handles image capture and computer-vision processing so that recognition happens locally and quickly; the cloud stores the resulting attendance records and serves them to a dashboard.
Keeping vision on the edge avoids streaming raw classroom video to the cloud, and keeping records in the cloud makes them available anywhere without trusting a single machine.
Architecture
- Edge (Python + OpenCV): captures frames and detects student presence, turning each session into structured attendance events.
- Data (Supabase / PostgreSQL): stores attendance logs and student records as the single source of truth.
- Dashboard (Next.js): a panel for instructors to review attendance, spot absenteeism trends, and confirm records.
Implementation
The core loop converts a live classroom into clean rows: detect, match, log. The instructor panel reads those rows back and renders per-student history and trends, so the output of the vision pipeline is always something a human can verify rather than a black box.
Challenges
The hard part was trust, not detection. Automated logs are only useful if people believe them, so a large share of the work went into validating automated attendance against manual records — comparing runs, checking edge cases, and confirming the panel displayed updated student data without visual glitches before relying on it.
Outcome
CObot+ automated attendance end to end and cut manual data entry by roughly 80%, with logs verified against manual records to confirm accuracy. Instructors gained a live view of absenteeism trends instead of a stack of paper sheets.
Lessons Learned
Automation earns adoption through verification, not cleverness. The detection was the interesting engineering, but the reconciliation work — proving the numbers matched reality — is what made the system something people would actually use.