Ziyi Li
← All work

03 / Tools for learning

Clear rules. Useful feedback.

A classroom coding workflow where flexible output rules still lead to repeatable results.

My contribution
Student project · product design & full-stack development
Status
Implemented student project
Focus
Next.js / FastAPI
Demo student assignment screen with visible decimal examples and a Python upload control.
Student workflow · archived interface with explicitly marked demo data.

The question

Make the rules flexible. Keep the judgment repeatable.

A correct program does not always print an answer in exactly one format. I built a classroom AutoGrader around that practical problem: teachers define what an acceptable answer means, and students receive feedback from a repeatable evaluation.

My contribution

  • Designed the teacher-to-student workflow: author a task, validate its checker, publish, submit and review.
  • Built a Python-only judging architecture with queued jobs, isolated execution and versioned checker specifications.
  • Separated optional AI-assisted checker drafting from deterministic student judging and explicit teacher publication.

How it works

Define and validate

Teachers choose exact text, token sequences, numeric tolerance or regular-expression captures. Accepted and rejected examples check a rule before it becomes active.

Publish a version

A checker draft must pass validation and be explicitly published. Optional model assistance drafts a specification; it does not decide a student’s result.

Submit and explain

Python files enter a queue and run in an isolated container. Feedback separates visible examples from hidden tests so the explanation does not disclose the test set.

Demo student assignment: decimal input, visible example and Python submission control.
Student view · read the problem, inspect visible examples, submit a Python file. Demo data.
Demo teacher editor with checker rules and accepted and rejected validation examples.
Teacher view · validate accepted and rejected examples before publication. Demo data.

An important change

The final product narrows earlier broad ideas into a Python-only, deterministic evaluator. AI moves to the teacher’s authoring step, where its output can be inspected and tested before publication.

Evidence & boundaries

The screens come from the project’s recorded demo workflow. The final packaged implementation and documentation match these interfaces. The portfolio does not connect to the application or its database.

Limits & next steps

The project is designed for a managed school intranet. Real deployment requires the sandbox, resource limits and host configuration to be verified together. Schoolwide adoption and production uptime are not claimed.