In April 2025, Ye Xu and Gary Leydon from the Yale School of Medicine (YSM) Educational Technology & Innovation team, and Lei Wang from the Harvey Cushing/John Hay Whitney Medical Library (CWML), met to learn about potential areas for collaboration. The trio immediately recognized a common goal: to streamline access to educational resources across YSM.
By the end of the meeting, Xu, business analyst lead, Leydon, associate director of medical education technology, and Wang, head of technology and innovation at CWML, had conceptualized an advanced curriculum search tool that would allow users to search through thousands of PDF files in just a few seconds. Now five months later, the team is rolling out the tool to YSM users.
Curriculum Search uses a multimodal large language model (LLM) and a retrieval-augmented generation (RAG) system to search thousands of files and find information about the MD curriculum.
Using materials such as lecture notes, presentations, and syllabi, the tool returns quick and accurate results for questions like “Do first year students learn about clonal hematopoiesis?” or “Can you give an overview of the end of life care in the curriculum?” and even “What limits the effectiveness of vaccines against malaria?”
“The medical school curriculum includes massive amounts of information,” notes Jaideep Talwalkar, MD, associate dean for educational technology and innovation at YSM. “Our tool makes this information more accessible and holds promise for a range of essential tasks, like mapping the curriculum to educational competencies, creating high-quality multiple-choice questions linked to course materials, and helping students find where specific topics are covered.”
The technology powering Curriculum Search is distinct from artificial intelligence (AI) tools like ChatGPT and Microsoft Copilot, which access a variety of data sources and are designed for general tasks like writing emails, coding, and generating images. Instead, Curriculum Search leverages a RAG system previously developed by Wang to restrict what sources the AI can consult when answering a query.
Wang, lead developer and designer of Curriculum Search, established the RAG system while building the Medical Library’s virtual assistant, ‘Ask Harvey’, to limit the chatbot’s data sources to the Yale Library catalog, Yale webpages, and PubMed. This ensures library users receive the most relevant and accurate information for their query. Similarly, the developers designed Curriculum Search with the same technological ‘gates,’ and it can only provide answers related to the MD curriculum materials it consults.
The development process took place over five months and involved several rounds of user experience (UX) testing to ensure the tool would meet the needs of its intended audience.
The team granted early access to 15 faculty members and a select group of staff involved in curriculum development. The creators asked early users to run controlled queries, report on the accuracy of results, and compare them against the source materials. The team also monitored the tool’s performance during the pilot phase and adjusted response parameters based on observed gaps or inconsistencies.
UX testing also highlighted a need for certain features. For instance, faculty wanted to know where the tool found the information and requested the addition of a citation list with links to the lecture notes, slides, lab materials, or reading assignments. Wang quickly implemented a source transparency feature to help users better understand and trust the tool’s responses.
Feedback also reflected a need for dynamic, conversational guidance to help construct questions that led to more specific answers. In response, the development team introduced lightweight prompt suggestions to assist in the querying process. Future enhancements include the ability to generate quiz questions and flashcards from lecture content or course material. Currently, the tool searches documents from the 2024 academic year. In the future, the content will update twice a year—in January and July.
Jeremy Moeller, MD, MSc, associate dean for curriculum in medical education, and early user of Curriculum Search remarked, “During our curricular redesign process, we have made extensive use of the tool to efficiently yet thoroughly identify how we are meeting our learning objectives, allowing us to reduce unplanned redundancy, create new sessions to fill gaps, and ensure that our teachers and course directors better understand how their material fits into the continuum of the curriculum. Moving forward, there is a huge opportunity for students and teachers to use the tool themselves to create personalized content and fine-tune their teaching and learning strategies."
In terms of security, Curriculum Search leverages built-in safety mechanisms afforded through Yale’s enterprise cloud platforms’ LLMs. The tool restricts use to campus internet protocol (IP) ranges and requires authentication through Yale’s central authentication system (CAS). The tool also includes a filter designed to detect and lock potentially harmful, malicious, or otherwise inappropriate prompts, as well as any attempts to bypass safety guardrails or manipulate the model.
The tool’s launch will follow a phased approach, with an expanded pilot coming in November 2025, and a full launch scheduled for January 2026. The full launch will include training sessions and onboarding materials to help new users effectively explore and use the tool.