Skip to content

How GradusMind uses Claude

GradusMind's AI features are built on Claude, by Anthropic. We use it where a student or teacher is waiting: grading a coding answer, explaining the moment a student is stuck, and running practice interviews. A teacher always makes the final call.

Start free trialBook a demo
  • Grades coding answers by running them
  • Explains a lesson when a student is stuck
  • Runs practice interviews with feedback

Grading that runs the code.

Before GradusMind's AI grading, coding answers waited for an instructor to mark them by hand. Now every submission gets a tested, explained score, and the instructor reviews instead of starting from zero.

  1. 1Student submitsA coding answer arrives from the in-browser editor with the question, its test cases and the instructor's rubric.
  2. 2Claude runs the codeUsing Anthropic's code execution tool, Claude runs the answer against every visible and hidden test case. Correctness comes from running the code, not from reading it.
  3. 3Scores and explainsClaude returns structured output: a result per test, a score and comment per rubric criterion, and feedback the student can act on.
  4. 4Instructor decidesScores are checked and totals computed in our code. The grade stays a suggestion until the instructor approves or edits it.
Grading resultSample, shortened
{
  "ranCode": true,
  "testResults": [
    { "testCaseId": "t1", "passed": true, "actualOutput": "8", "error": "" },
    { "testCaseId": "t7", "passed": true, "actualOutput": "-1", "error": "" }
  ],
  "rubric": [
    { "criterionId": "correctness", "score": 38,
      "comment": "All visible and hidden test cases pass." },
    { "criterionId": "edge-cases", "score": 6,
      "comment": "Duplicates handled; empty array not checked." }
  ],
  "feedback": "Clean single pass. Guard the empty input.",
  "suggestions": ["Return early when nums.length < 2"]
}

Scores are clamped to each criterion's maximum and totals are computed by GradusMind, never taken from the model.

A tutor that notices when you're stuck.

When a student rewinds the same part of a lesson again and again, GradusMind offers help. Claude explains that exact moment from the lesson's transcript and what's on screen, in Hindi or English, and answers follow-up questions.

Faculty see which minutes of a lesson confuse students most, so they know what to teach again.

Spring Boot REST APIs, lesson 412:30 to 13:00

Looks like this part is tricky. Want a quick explanation of @PathVariable vs @RequestParam?

Student asks

Why does my URL need curly braces here?

Claude explains

The braces mark a part of the path that changes, like a user's ID. Spring fills it into your method with @PathVariable.

Sample conversation.

Why we build on Claude.

  • Code execution

    Claude runs student code in Anthropic's sandbox, so a grade reflects what the program actually does.

  • Structured outputs

    Every grade comes back in a fixed schema the platform can validate, store and show without parsing prose.

  • Long context with caching

    A full lesson transcript fits in one request, and prompt caching keeps repeat questions on the same lesson inexpensive.

  • Hindi and English

    Students ask in the language they think in, and get explanations at their level.

  • Reads what's on screen

    A frame of the slide or code a student is stuck on can be included, so the explanation matches what they see.

  • Safety built in

    If a request is declined by a safety check, for example exploit-like code, it is retried on another model rather than failing silently.

Teachers stay in charge.

AI should save teachers time, not take decisions away from them. These rules are built into the platform.

  • Every AI grade is a suggestion until an instructor approves it.
  • Each institute can turn AI features on or off and set monthly limits.
  • Per-student rate limits keep usage fair and costs predictable.
  • Token usage is logged for every request.
  • Only what a feature needs is sent: the question, the answer or the lesson moment, never unrelated profile data.
  • Anthropic's commercial terms do not allow it to train models on data we send.

Next on our roadmap

Early warning for students falling behind

A weekly summary for faculty of who is slipping on attendance, progress and scores, prepared overnight with Claude's Batch API so it costs less.