Problem
Feedback workflows need consistency and speed, but the system cannot substitute model output for teacher judgment or expose private instructional records.
Project Brief
Instructional AI Workflows
LMS-agnostic workflow prototypes for rubric-defined evaluation, draft feedback generation, human review, and student-facing artifacts.
Overview
Teacher-controlled AI assistance
The project frames AI as workflow support, not automatic grading: teacher-defined rubric -> structured evaluation -> draft feedback -> human review -> student-facing feedback, review, or remediation artifact.
Current Artifacts
- Precalculus FRQ synthetic rubric demo
- Three synthetic learner evaluations
- Two review-required feedback examples
- Structured evaluation JSON
- Teacher reviewer packet
- Reviewed student-facing feedback
- Remediation planning output
- Public safety rules for synthetic submissions and rubrics
System Story
From rubric evidence to reviewed language
Approach
JSON rubrics and synthetic responses feed deterministic workflow logic that separates evidence capture, draft generation, teacher review, release language, and remediation planning.
Result
The Precalculus demo produces three synthetic evaluations, two review-required cases, a reviewer packet, approved feedback, and a rubric-organized remediation plan.
Lesson
The strongest use of AI here is a controlled review pipeline, not automatic scoring. Evidence and teacher approval remain explicit at every release boundary.
Demo Evidence
What the demo shows
Synthetic FRQ input
The source repository includes fake learner responses and a teacher-defined rubric for a Precalculus linear model task.
Structured evaluation
The demo produces rubric-level evidence, scores, private teacher notes, and review-required flags for three synthetic learner examples.
Reviewed output
Only approved release notes are shown as student-facing feedback; teacher-facing review details remain separate.
Remediation planning
The demo groups rubric evidence into teacher-facing next steps for model setup, solving process, and context interpretation.
Safety Boundary
No private instructional records
Public demos must use synthetic submissions, synthetic rubrics, fake course identifiers, and public-safe sample outputs. Do not publish real student submissions, feedback, grades, comments, Canvas IDs, assignment IDs, course IDs, API tokens, or teacher-only review artifacts.