This abstract was part of the 2026 UCSF AI and Education Symposium
Undergraduate medical education offers few dedicated opportunities to practice clinical reasoning. The most widely used medical school learning platforms like UWorld and Anki, as well as pre-clinical pedagogical tools such as inquiry-driven worksheets, emphasize bottom-up learning by building disease-level pattern recognition and factual recall to support clinical decision-making. However, these skills do not consistently translate to real-world clinical performance. When an undifferentiated patient presents to the emergency department, the neatly organized symptom constellations encountered in question stems or flashcards present limited utility without a structured framework for generating a differential diagnosis and assessing disease severity. Opportunities to develop these frameworks occur late in undergraduate medical education and require substantial faculty engagement or curricular time.
However, studies of master clinicians suggest that the development of clinical excellence relies heavily on sustained, self-directed learning. Educational forums such as morning report, morbidity and mortality conferences, and case reports currently serve as the primary mechanisms through which clinicians continually refine their clinical decision-making. Earlier access to similar instructional methods could enable learners to develop framework-based approaches to pathology, diagnosis, and management sooner and more systematically. Generative AI offers a unique opportunity to support this shift by synthesizing high-quality clinical cases at scale and delivering structured, feedback-driven guidance on diagnostic reasoning and management, while reducing faculty burden.
Here, we propose MorningReportAI, an AI tool that can be used by preclinical trainees to build frameworks and approaches to clinical scenarios.
MorningReportAI is grounded in the principle that any clinical learning scenario can be systematically decomposed into four core stages: (1) approach to the chief complaint, (2) diagnostic evaluation, (3) data interpretation, and (4) management. At the chief complaint stage and diagnostic evaluation stages, the system intentionally constrains learners to identify only three pieces of pertinent information and diagnostics, encouraging prioritization and hypothesis-driven reasoning. This structured approach also introduces an important equity dimension by preparing learners to reason effectively under real-world constraints. In the final stages, learners synthesize available data to arrive at a diagnosis, propose a management plan, and articulate a single key takeaway, reinforcing reflection and knowledge transfer. MorningReportAI also provides a platform for educators to contribute educational case reports or curated case summaries for learner use. This creates a continuous feedback loop of real-world clinical cases, enriched by faculty-annotated approaches and key takeaways, that can be iteratively reused and expanded for education.
Please see this video for a demo: https://youtu.be/-20hyoD-ze4.
The effectiveness and accuracy of the platform can be evaluated through multiple metrics. First, educator engagement can be quantified by tracking the number of cases contributed for student learning. Second, learner interaction and educational value can be assessed by monitoring user responses to cases and the quality of documented learning takeaways, all of which are captured by the platform. Finally, MorningReportAI incorporates a retrieval-augmented generation (RAG) framework for clinical approaches and employs large language model–based validation checks to ensure the accuracy and reliability of generated content. Ultimately, MorningReportAI has the potential to revolutionize preclinical medical education by advancing clinical reasoning skills at the earliest stages of training.
Contact
Akash Shanmugam, [email protected]