doerkit: An open-source LLM-graded assessment platform with spaced review

Short summary

The author releases doerkit, an open-source assessment-first course platform that uses an LLM to grade written answers against rubrics rather than acting as a chatbot tutor. The system wraps grading in spaced cumulative review, which carried the largest effect size in the source study. Two MIT-licensed repos are provided: doerkit (platform) and rubric-bench (regression testing for LLM judges). The author honestly catalogs gaps between demo and production, including LMS integration, FERPA compliance, and the need for an RCT.

• •doerkit is an open-source LLM-graded assessment platform with spaced review, not a chatbot tutor
• •rubric-bench provides regression testing for LLM judges including adversarial prompt-injection suites
• •Author transparently lists gaps: LMS integration, FERPA compliance, inter-rater validation, and RCT needed

Generated with AI, which can make mistakes.