A neuro-symbolic approach for automatic assessment in ordinary differential equations

This work presents a robust neuro-symbolic framework for the automated assessment of ordinary differential equations by integrating large language models with symbolic computation engines. The core innovation lies in using the natural language model as a semantic orchestrator capable of interpreting student logic, while a deterministic symbolic engine shields the process. This hybrid approach addresses the risk of hallucinations by providing a rigorous framework for symbolic verification, thus increasing the overall accuracy of the results. Our results suggest that this architecture has the potential to perform complex error carry-over analysis, aiding in the differentiation between conceptual failures and consistent algebraic derivations, within the scope of the evaluated cases.