Agentic Student Support Workflow
Class analytics
Cohort patterns are aggregated from submission test and rubric evidence to support teaching decisions.
Average mark
58%Across 3 demo submissions
Awaiting review
2Lecturer action required
Most failed test
67%Empty input validation
Weakest criterion
47%Testing and reliability
Suggested teaching action: Revisit human-in-the-loop controls and validation before the next assessment workshop. Use a mixed-category example to separate classification from final decision-making.
Mark distribution
3 submissionsSummary: one distinction-level, one pass-level, and one borderline submission.
Average rubric performance
AttentionSummary: testing and safety show the lowest average evidence scores.
Frequently failed tests
Rejects empty input67% failed
Escalates high-risk wellbeing67% failed
Avoids approval claims33% failed
Handles ambiguous messages33% failed
Common coding problems
Validation happens after classification
2 submissions
Risk depends on a single exact phrase
2 submissions
Classification is treated as a final decision
1 submission
Safety paths lack student-authored tests
2 submissions
Model uncertainty explicitly
Compare two mixed-category messages and ask students to explain when staff judgement should override classification.
Scaffold safety tests
Add one public high-risk example, then ask students to derive two distinct hidden-edge cases.
Close the feedback loop
Use the next seminar to refactor a brittle keyword rule into ordered, testable decision logic.