grading-student-grading

Automated per-student grading launches an independent sub-agent for each student, enabling parallel assessment while preserving per-student isolation and easy resume capabilities. It processes extracted answers against a rubric and outputs individual results.

Yes, the grading workflow is resume-safe. It writes output per student to individual files, allowing the process to re-grade or continue from where it left off without duplicating completed work.

Sub-agent isolation ensures each student is graded independently, preventing cross-contamination of results. Model consistency is maintained by sharing the main agent's model across all sub-agents, ensuring uniform grading standards.

Required inputs include a grading guide, each student's extracted answer, and any relevant page images. Optional reference calibration can be added to improve grading accuracy and align with assessment standards.

Concurrency is controlled by a configurable BATCH_SIZE parameter. This regulates how many sub-agents run simultaneously, balancing grading speed against system resource limits for large-scale student assessments.

Yes, the workflow validates outputs before writing. This ensures only properly formatted grading results are saved to the per-student output files, maintaining data integrity across the entire batch.