The field of educational assessment is undergoing significant transformations with the integration of large language models (LLMs). Recent developments indicate a shift towards more nuanced and multifaceted evaluation methods, moving beyond traditional holistic assessment approaches. This trend is characterized by the incorporation of pedagogical theories and frameworks, such as Vygotsky's theory and Bloom's Taxonomy, to inform the design of LLM-based assessment tools. These innovations aim to provide more accurate and comprehensive evaluations of student performance, addressing the complexities of cognitive development, disciplinary thinking, and academic competencies. Noteworthy papers in this area include PEMUTA, which pioneers a pedagogically-enriched framework for multi-granular undergraduate thesis assessment, and SketchMind, which introduces a cognitively grounded, multi-agent framework for evaluating student-drawn scientific sketches. Additionally, ELMES provides an automated framework for assessing LLMs in educational settings, while CoGrader and DUET demonstrate the potential of human-LLM collaborative grading systems and LLM-based tools for providing structured feedback on student-generated diagrams.