A cognitively-grounded multi-agent framework that addresses scoring consistency and interpretability for open-ended student responses, leveraging Bloom’s Taxonomy-based Teacher Agents, RAG, and DPO.