Open-response Grading Assistant

The problem this kills

Grading scenario, essay, and case-based assessments is the part of training that quietly eats your week. You read the same question fifty times, you try to apply the rubric the same way to answer #3 and answer #43, and by the end you are tired, inconsistent, and behind. So you either rush it, water down the rubric, or stop using rich open-response questions at all and fall back to multiple choice that tells you nothing.

The usual "fix" is to let an AI just grade everything. That is worse: you hand the judgment call to a machine, the score is a black box, and the first time a learner challenges a grade you have nothing to stand on.

This Implementation Plan gives you the middle path that actually works: the AI does the tedious first pass against your rubric, and you stay the grader of record.

What you'll build

A private internal tool that:

• Stores your open-response questions and a scoring rubric for each one.
• Takes in submitted answers (typed in, or imported from a spreadsheet).
• Has the AI draft a score and written feedback for each answer, keyed to your rubric criteria.
• Shows you each suggestion side by side with the answer and the rubric, and tells you which criteria the AI keyed on.
• Flags low-confidence or out-of-scope answers so you look harder at the ones that need it.
• Lets you edit the score and feedback and approve the final grade. Nothing is recorded on the AI's score alone.
• Records the approved grades, notifies each learner, and exports a clean grades CSV.

What's inside the Implementation Plan

The plan is a single file you paste into Claude Code (a free AI coding tool). It builds the whole app step by step, and each step ends with a ready-to-copy prompt.

The most important part: the plan opens by interviewing you about your business. Before it writes a line of code, the agent asks how you run assessments today, what your rubrics and scoring scales actually look like, how learners and questions are named and identified, your typical and peak grading volumes, your approval rules, and your messy edge cases. Then it reads a short tailored spec back to you for a thumbs-up. You get a tool shaped to how you actually grade, not a generic template you have to bend to fit.

Inside you'll find: the discovery interview, the data model, the AI drafting step with rubric-criteria transparency, the trainer review-and-approve screen, learner notifications, and the CSV export. Plus a "How to know it works" checklist.

The governance it includes (this is the point)

This is built to be trusted with real grades:

• Login so only your team can get in.
• Row-level security so each organization only ever sees its own assessments and learners.
• A human-in-the-loop approval gate so the AI only ever drafts. A trainer reviews, edits, and approves before any grade is recorded.
• A full audit trail of who approved what and when, including the AI's original suggestion versus the final grade.
• Duplicate guards keyed on learner plus question plus submission, so the same answer can never be graded or recorded twice.

Who it's for

Trainers, instructional designers, and enablement leads who run scenario, essay, or case-based assessments and currently spend hours grading long answers by hand, want consistency across a batch, and need to stay the human in charge of every grade.

You've got this. Paste the first prompt and let the interview tailor the rest.