Automated Marksheet Grading
End-to-end exam evaluation from scanned marksheets to Excel report
RoleDesigner & sole engineer
PeriodMar 2025 — Apr 2025
StackPython, YOLOv8, CNN, Next.js
Categoryml, automation
Problem
Manual marksheet grading at scale is slow and error-prone.
Approach
- Trained a YOLOv8 model on a custom-annotated dataset to detect student identifiers and per-question marks
- Built a CNN for handwritten digit recognition on top of the detection pipeline
- Built a Next.js frontend for bulk uploads that returns downloadable Excel reports
Pipeline
Scanned upload
YOLOv8 detection
CNN digit recognition
Excel report
Outcomes
Manual time saved~75%
Identifier/mark detection accuracy~90%+
Digit recognition accuracy~92%