ProofPaths: An AI Agent Framework for Assessing Student Reasoning
| Source: arXiv AI
Tags: ProofPaths, AI Grading, Multi-Agent Systems, Education Technology, EduHiPC
ProofPaths pairs multi-agent rubric grading with AI-led student interviews that ask learners to justify steps in their submissions, aiming to check understanding when LLMs can write the answers.
Details
Students can hand in correct LLM-written solutions without understanding them, and one-on-one interviews that check understanding take recurring staff time. ProofPaths combines rubric-based grading with interactive interviews. Several grading agents assess a submission against instructor guidance, critique each other, and revise; their scores pick which review supplies the grade and feedback. An interview agent asks students to justify steps and adapts follow-ups to their answers. Each question links to a retrieved excerpt of the submission so instructors can compare explanation and work. Instructors can inspect all reviews and transcripts and override grades. The paper was accepted at the EduHiPC 2026 workshop and reports deterministic tests of the assessment controllers rather than a classroom evaluation.