From AI Flashcards to XR Anatomy: YSM Teams Build Prototypes to Transform Medical Education

A clerkship scheduler that places 100% of students in their first or second rotation choice. An artificial intelligence (AI) tool that converts specialty guidelines into targeted flashcards. A clinical coaching app that gives trainees structured feedback on empathy and communication.

These were among the working prototypes that emerged from YSM’s second Ideathon on April 20, where faculty, staff, students, and residents spent an afternoon using AI and extended reality (XR) to collaboratively tackle challenges in medical education.

Jaideep Talwalkar, MD, associate dean for educational technology and innovation, and Kathleen Ludewig, MPP, MSI, IT director and co-lead of YSM Educational Technology & Innovation, gave opening remarks. They highlighted several successes from last year’s Ideathon and encouraged the 33 attendees to spend the next few hours experimenting, iterating, and learning together.

“Today we invite you to connect with people from different backgrounds and skills to look at these ideas with new perspectives,” said Ludewig. “We are strict about time limits for activities but everything else is open. You have full creative control.”

Attendees heard from six individuals with unique challenges and learned about Yale AI tools that they could use to facilitate ideation and product development. Following the presentations, attendees were asked to break into small groups to brainstorm solutions. Of the six presentations, four were selected for further development.

Michail Kokkorakis, MD, PhD, postdoctoral fellow in the section of digestive diseases, proposed an AI-powered flashcard system to support memorization of specialty guidelines.

The problem

Each medical specialty follows a set of society guidelines containing evidence-based recommendations for standardizing care and optimizing patient outcomes. In general, these guidelines are lengthy and densely written, which can pose challenges in terms of recall and comprehension.

While flashcard tools such as Anki and Quizlet can help with retention, they require significant time to set up.

The idea

Kokkorakis proposed creating an AI-powered platform that would take these guidelines and extract key recommendations, decision points, drug doses, and diagnostic criteria. The AI would then generate flashcards based on the content, tag it by topic and difficulty, and include a citation to the page and guideline. As a bonus, the AI could schedule strategically spaced review times to ensure optimal learning and retention.

Unlike Anki, these flashcards would come directly from the guidelines instead of crowd-sourced content.

The proposed solution

During the Ideathon, Kokkorakis collaborated with Sam Friedman, a computational research support analyst from Yale Center for Research Computing, to build a prototype that generated high-quality, accurate flashcards from a limited set of PDFs. Going forward, Kokkorakis plans to compare the working prototype against traditional methods of learning to measure its efficacy. He also plans to seek broad feedback and set up a small usability study.

Danette Morrison, medical student clerkship coordinator, presented attendees with a complex scheduling challenge with many variables.

The problem

Scheduling clerkship blocks is a time-consuming, manual, and error-prone process. It involves scheduling 20 to 30 students for four three-week rotations, choosing from a catalog of more than 20 core rotations and sub-specialties, while honoring individual student preferences and adhering to rotation capacity limits. Currently, the process is organized on a complex spreadsheet which tracks each student’s clerkship block, the dates and times for each rotation, and their preferences and special considerations.

The idea

Morrison sought attendees’ help in creating a platform that would automatically match and optimize student rotations across each clerkship block. Ideally, the tool would include a dashboard that displayed conflicts and would reoptimize scheduling when a student’s needs or preferences changed. An additional ask, or “bonus challenge” as Morrison called it, was the ability to schedule call nights, a process with even more variables.

The proposed solution

During the brainstorm, Morrison’s group used Morrison’s spec sheet, Excel files, and one semester of data to create a prototype. With help from Tristen Lawrence, a software engineer from ITS Cloud Engineering, they built a working PHP prototype in less than an hour.

The Clerkship Scheduler has a dashboard that shows the number of students, rotations, and assignments, as well as the percentage of students who get their first or second choice for rotations. The model placed 100% of students within their first or second choice and evenly assigned students to mandatory rotations across the 12-week schedule.

When asked what’s missing from this prototype, Morrison responded “Nothing, it’s actually what I dreamed of! In the future, I hope everyone will be able to use AI to produce solutions like this without requiring a software engineer.”

Saeed Juggan, MD-PhD candidate, presented an opportunity to bring photogrammetry-generated 3D anatomical models and radiology imaging together using Apple Vision Pro headsets and the Visage Ease Pro app.

The problem

All medical students must demonstrate proficiency in human anatomy, but it can be challenging to translate anatomical structures to what’s captured on radiology images. This leaves students feeling ill-equipped to read and understand scans. Recently, YSM students have been able to explore dissected and pre-dissected 3D anatomical specimens using Apple Vision Pro headsets. Combining these 3D anatomical specimens with corresponding radiographic imaging could significantly improve understanding and contextualization of what they see.

The proposed solution

Juggan proposed collaborating with experts in radiology, anatomy, and commercial partners to develop and test a spatial computing platform that uses Apple Vision Pro and Visage Ease Pro. The platform would integrate radiographic images and 3D photogrammetry models. Juggan and his team are working with Visage Ease Pro to combine radiographic images and Visage’s 3D models with YSM’s photogrammetry models.

Next steps

To evaluate the educational effectiveness of this tool, Juggan and his team will recruit 20 student volunteers enrolled in the Human Anatomy course. Volunteers will use Apple Vision Pro headsets to interact with 3D anatomical models and radiology images through the Visage Ease Pro application. The team will collect qualitative and quantitative data from each student and administer a quiz to assess knowledge retention and the tool’s impact on learning outcomes.

Darice Corey, senior director for web & IT planning at Yale College, proposed an AI model that provides structured coaching for trainees in clinical settings.

The problem

After reviewing published studies, Corey found that medical trainees do not always receive consistent or timely feedback after patient encounters, and that feedback often focuses on clinical accuracy rather than skills such as tone and empathy. This, she reasoned, can lead to delays in skill development and inequitable learning experiences among trainees.

The idea

To improve feedback and skill development, Corey proposed an AI-powered clinical coaching tool that offers consistent, structured feedback for all trainees. The tool provides foundational feedback in four areas (clinical reasoning, clarity of communication, empathy and tone, and equitable practice) and enables faculty members to focus on more qualitative observations.

The proposed solution

Corey built a clinical coaching prototype in Replit AI that allows users to upload an audio or text transcript of a clinical encounter and receive feedback. The prototype enabled the group to test the concept and address questions about data protections, capturing non-verbal clinical competencies, and how feedback should be conveyed (through numeric scores or narrative summaries). The group also discussed what would be needed to move the prototype into production, such as encrypted data storage, secure authentication, role-based access, session recording, and tracking progress over time.