Lecturer review required

John Brown / Agentic Student Support Workflow

14 of 16 tests passed / Evidence completeness 94%

Provisional result

AI-generated evaluation is advisory. The lecturer reviews the evidence and makes the final academic decision.

87%

Provisional overall mark

Distinction-level evidence profile

14 tests passed 2 tests failed

Functional correctness

35% weighting / 30.8 marks

88%

Agentic workflow and orchestration

20% weighting / 17.2 marks

86%

Safety and human oversight

15% weighting / 13.5 marks

90%

Testing and reliability

15% weighting / 11.7 marks

78%

Code quality

10% weighting / 8.8 marks

88%

Documentation

5% weighting / 4.6 marks

92%

Strengths

Clear separation between classification, risk assessment, action planning, and response drafting.
All high-risk cases trigger human review and drafts avoid promising outcomes.
Meaningful unit tests cover validation, classification, and safety paths.

Areas for improvement

Ambiguous messages containing both technical and wellbeing signals are classified by first match.
One edge case and one uncertainty-handling test failed.

Code-level observations

src/index.ts orchestrates four focused functions with typed outputs.
risk-assessor.ts handles immediate-danger phrases before category-specific rules.
README.md documents setup, design decisions, and human-review boundaries.

Editable draft

Feedback to student