Assessment Paper Engine

A generate-then-verify agent system with an automatic repair loop.

One agent computes every answer by executing Python, not recalling it. A second agent re-solves the whole paper blind, in a fresh context, with no view of the first agent's working — and any disagreement sends it back for regeneration automatically. Nothing reaches a student without a passing validation report.

System Architecture

Architecture diagram

Coming soon

Technology Stack

Claude Agent SkillsPython + SymPyFresh-context subagentBounded repair loopChapter knowledge filesPDF generation