Objective: This study aimed to identify the dimensions and components of intelligent learning assessment in primary education based on artificial intelligence.

Methodology: This was an applied study with a qualitative nature and an exploratory approach. The statistical population consisted of experts in the fields of educational sciences, educational assessment, and artificial intelligence, selected via purposive sampling until theoretical saturation was reached. Data were collected through in-depth, semi-structured interviews and analyzed using three stages of coding: open, axial, and selective. Holsti’s coefficient of agreement was used to assess reliability.

Findings: In the open coding stage, 1,078 initial codes were extracted and subsequently refined to 420 open codes. The Holsti agreement coefficient between the two coders was calculated at 0.86, indicating satisfactory research reliability. Ultimately, the study's final structure was defined across five macro-dimensions: paradigm shift (transition to a process-oriented approach), design (individualized adaptation), implementation requirements (technology governance), data-driven analysis (technological integration), and feedback (decision-making for new actions).

Conclusion: The identified structure and dimensions represent a comprehensive, multi-layered system for intelligent assessment. This system can serve as a strategic foundation for educational policymaking, the design of intelligent assessment platforms, teacher empowerment, and the development of future studies in this field.