Capstone Proposal · Planning Stage

Work Submission & Grading System

An AI-graded pipeline (Gemini API) for MC, FRQ, and GitHub Issue submissions, storing every score and feedback the way the existing game-submission spreadsheet does today, in a database admins can browse.

Team: Nicolas Diaz, William Windle, Rishabh Jha · 2026-2027

🚧 Research Question, Design Decisions, System Architecture, and Progress are still to be continued — nothing above this line has been built yet.

What We Plan To Do

AI feedback on game submissions currently lives in a spreadsheet: someone pastes in the submission, the AI grades it, and the result gets typed back in as a row. We plan to replace that with a real database, and add two new submission types (FRQ and GitHub Issue) alongside the existing MC type, all graded through the Gemini API.

Reference material from the planning issue

1 · The spreadsheet this replaces

Current game-submission feedback spreadsheet — student, AI grade, and AI feedback columns are what the new database needs to hold.

2 · The system this builds on

The existing frontmatter-driven MC/FRQ submission system that GitHub Issue submissions will be added alongside.

3 · The rubric the AI will follow

Grading definitions from the spreadsheet — the basis for the AI grader's rubric.

Submission types

Type	How it's graded
📌MC	Selected answers, checked against an answer key by Gemini.
✍️FRQ	Written free response, graded by Gemini against a rubric.
🔗GitHub Issue	Link to a public issue, scraped and sent to Gemini for a dynamic review — for more complex tasks.

Types are defined through frontmatter, the same pattern the site already uses. For GitHub Issue submissions: the link must be publicly viewable, it's checked for validity/access first, then formatted and sent to the grader. A GitHub API token will be used when available; Gists may hold submission content where needed.

Storing results

Every graded submission — student, type, AI score, AI feedback, rubric version, and timestamp — gets written to the database instead of a spreadsheet row, so admins can browse submissions directly rather than maintaining a sheet by hand.

Teacher requirements

!Every AI-generated comment must include a clearly marked "AI" disclosure tag.

!A GitHub Issue documenting exactly how the grader is implemented must be finalized alongside the build.

!Both batch grading and individual grading need to be supported.