Pilot Projects

Below is a partial list of pilot projects selected for AI experimentation following a Call for OpenAI Proposals offered in Fall 2024.

Executive Summary

The promise of inclusive education remains largely unfulfilled as faculty training on accommodating students with disabilities continues to fall short. A recent cross-sectional study on average rated the top 50 NIH-funded undergraduate programs a 1.2 out of 10 in terms of accessibility and disability inclusion (Swenor et. al.). This proposal leverages OpenAI to develop a cloud-based platform in which faculty and students can receive real-time guidance on disability-specific accommodations. The first aspect will involve providing feedback to faculty members on implementing universal design learning in their courses, accessibility audits, and specific resources for individualized scenarios. The second aim is to provide neurodivergent students with various command prompts to receive personalized solutions ranging from course-specific study schedules to learning tips. In general, a data-driven decision-making approach will allow for streamlined information to accommodate services for faculty and students. This initiative promotes equity, diversity, and inclusion in optimizing every student’s education, a limitation of UCLA’s current 1:400 specialist to disability-student ratio. In hopes of fostering intercultural fluency, UCLA can provide the first AI-driven guidance system specifically targeted to disability and accessibility topics, catalyzing the implementation of inclusive teaching practices in education.

Project Goals

This proposal centers on developing a dynamic, cloud-based platform that provides various resources to faculty and students. By feeding this platform with current web-based data, methodological procedures from the CAE department, and human-based experiences, an assistive source can be developed.

Specific Aim 1: The platform will equip faculty with personalized tools and resources to provide instant responses to individualized questions. Topics range from disability-specific accommodation guides and processing with the CAE, inclusive syllabus builders, accessibility checks of course material, and training on universal design for learning.

Specific Aim 2: By providing numerous command prompts to submit to the platform, students can receive personalized feedback for optimizing their academic experiences. This ranges from adaptive study schedules, interactive concept breakdowns, and personalized study tips that account for their disability.

This AI-driven accessibility innovation elevates the academic experiences of 15 percent of UCLA’s student population that affiliated with having a disability.

Duration

12/18/24 - 6/01/25

Project Lead and Team Members

Aryan Pandey (UCLA Senior)

Executive Summary

This proposal aims to develop an AI Agent grounded in UCLA graduate education policy, specifically focusing on the Codification of Policies and Procedures Governing Graduate Admissions, the Standards and Procedures for Graduate Study at UCLA, and Graduate Program Requirements. By training an AI Agent on these policies, we aim to elevate the UCLA student, faculty, and staff experience by providing a flexible, user-friendly, and efficient solution to improve access to critical graduate education information and decrease administrative burden.

Project Goals

This AI-driven accessibility innovation elevates the academic experiences of 15% of UCLA’s student population that affiliated with having a disability.

Duration

1/06/25 - 6/30/25

Project Lead and Team Members

Chris Testa (Director, Information Technology, Division of Graduate Education)
Vania Sciolini (Assistant Director, Strategic Communications, Division of Graduate Education)
Priya Thangaraju (Senior Application Developer, Division of Graduate Education)
Kristin Ashton (Senior UX Designer, Division of Graduate Education)
Reid Johnson (Project Manager and Business Analyst, Division of Graduate Education)

Executive Summary

With the growing popularity of AI applications in the public sector, many Humanities instructors have approached Humanities Technology seeking guidance on incorporating such applications into their classrooms. We propose using an enterprise AI license to develop and curate an "AI Toolkit" for classroom instructors and teaching assistants. This toolkit will include resources such as AI best practices in the classroom, sample AI-based activities and lesson plans, and AI-based assignments and assessments. The toolkit will be developed in collaboration with our Research and Instructional Technology Consultants (RITCs), graduate student employees focused on instructional and internal research support. Utilizing the AI license, the RITCs and I will develop these resources, test them with faculty volunteers, and make them publicly available on our HumTech website.

Project Goals

This project aims to develop and curate an “AI Toolkit” for Humanities instructors, enabling them to effectively incorporate AI applications into their classrooms. The toolkit will include resources such as AI best practices in the classroom, sample AI-based activities and lesson plans, and assignments and other AI-based assessments. The project will be developed in coordination with Research and Instructional Technology Consultants (RITCs), graduate student employees focused on instructional and internal research support. The primary goals are to:

Empower Humanities instructors to leverage AI in their teaching practices
Provide practical resources and guidance for implementing AI in the classroom
Enhance student learning experiences through innovative AI-based activities and assessments
Make the "AI Toolkit" publicly available via the HumTech website, promoting widespread adoption and collaboration among Humanities educators

The enterprise OpenAi license will be used to develop the educational materials, which will be complemented by a few volunteer courses to test their effectiveness. A successful project would include positive feedback from our case studies and the publication and dissemination of said toolkit on the HumTech website.

Duration

1/01/25 - 6/13/25

Project Lead and Team Members

Dr. Jordan Galczynski, Instructional Technology Manager
Anna Bonazzi (Ph.D. Candidate, German)
Jae Hyeon Park (Ph.D. Candidate, Urban Planning)
Tianji Jiang (Ph.D. Candidate, Information Studies )
Emma Ridder (Ph.D. Candidate, English)
Saad Shaukat (Ph.D. Candidate, NELC)
Anna Brungardt (Ph.D. Candidate, German)
Erika Snell, Instructional Technologist

Executive Summary

UCLA Xplore is an AI-driven chatbot designed to address the challenge of information overload and scheduling conflicts faced by students navigating extracurricular opportunities. Leveraging OpenAI's embedding and GPT models, the project will develop an extracurricular option recommendation bot that summarizes vast amounts of event data and provides personalized suggestions based on students' interests and schedules⁠⁠. The chatbot will integrate with students' indicated available time slots to avoid event clashes and enhance overall campus engagement⁠⁠. By streamlining access to campus events and resources, UCLA Xplore aims to improve existing student participation, optimize on-campus resource utilization, and showcase UCLA's vibrant campus life to a wider public⁠⁠. This scalable solution has the potential for significant long-term benefits in student engagement and university attractiveness.

Project Goals

Scope: the OpenAI licenses will be used to call
- OpenAI Embeddings : Power a recommendation system utilizing Retrieval Augmented Generation
- OpenAI 4o / 4o mini: Synthesize relevant activities options information
Objectives:
- Enhanced Student Engagement: Increase student extracurricular participation by providing personalized and accessible information.
- Optimized Resource Utilization: Improve attendance at on-campus events through targeted recommendations.
- 20+ hours custom support effort saved weekly by answering repetitive questions for Student Affairs customer support.
- Boost Prospective Student Interest: showcase UCLA’s vibrant campus life to prospective applicants and outsiders.
Outcomes / Deliverables:
- Personalized Event Recommendations: Students receive tailored suggestions based on their interests, academic schedules, and past engagement patterns.
- Schedule Integration: The chatbot syncs with students' class schedules to recommend events that fit their availability, minimizing scheduling conflicts.
- Real-Time Updates: Utilizes Retrieval Augmented Generation (RAG) to provide up-to-date information on campus events and activities.

Duration

2/01/25 - 6/01/25

Project Lead and Team Members

Wanxin Xiao, UCLA College of Letters and Science 2027
Hunter Kang, UCLA Samueli School of Engineering 2027
Krisha Basrur, UCLA Samueli School of Engineering 2027
Saatvik Sharma, UCLA Samueli School of Engineering 2027
Alexander Hu, UCLA Samueli School of Engineering 2027
Uzair Hammad, UCLA Samueli School of Engineering 2028

Executive Summary

The Emerging Research Trends Dashboard program is a proposal based in supplementing UCLA students' education by providing guidance through cutting-edge developments in their chosen field of interest. The dashboard will utilize Artificial Intelligence to analyze vast amounts of academic literature, research articles, and other sources to find trending ideas in academia and highlight them to the user to supplement their pursuit of developing a credible hypothesis.

Project Goals

The Emerging Research Trends Dashboard will utilize OpenAI licenses to analyze the vast datasets of academic publications and conference proceedings using natural language processing. This project aims to identify trending research topics, find influential papers, and provide students with personalized feedback based on their unique academic interests and career goals. Using AI-driven trend analysis, interactive learning modules, and collaborative features, the dashboard will aim to simplify access to new and exciting knowledge, allow students to explore niche topics, and connect them with relevant faculty and labs on campus. The expected outcomes include easier access of emerging fields, greater connections within the research community, and will offer better preparation for graduate studies or careers in innovative industries. If successful, the dashboard will simplify complex research areas and serve as a valuable resource for UCLA students to contribute to quickly evolving academic landscapes.

Duration

1/01/25 - 5/31/25

Project Lead and Team Members

Prakash Jeysankar, UCLA Molecular, Cell, and Developmental Biology Class of 2026
Aakash Jeysankar, UCLA Ecology and Evolutionary Biology Class of 2026
Jeysankar Ramakrishnan, Manager, Workplace IT Products

Executive Summary

Good provider-patient communication is vital to improve patient engagement in healthcare and outcomes. Standardized patients (SPs) prepare clinicians to communicate with patients, but SP training is constrained by available resources and competing clinical demands. Training often focuses on history taking and other common clinical encounters with less time devoted to breaking bad news as a vital skill. Researchers are turning to GPT and other LLMs to create communication skills simulations that incorporate computer-generated (virtual) SPs (VSPs). Our team has experience developing VSPs, most recently for a NCI-funded pilot study to help medical students learn to discuss the results of an abnormal mammogram with patients. We developed the VSP using Hyperskill, simulation authoring software that incorporates GPT-4 as a beta feature. While successful, we would like to streamline development by acquiring OpenAI ChatGPT licenses. Currently, we scripted training scenarios without AI assistance and could only test the VSP through Hyperskill software, which slowed down the development process. We are requesting OpenAI licenses to more rapidly develop communication skills training scenarios and prototype them using ChatGPT with the research team members.

Project Goals

The project will use ChatGPT to better develop communication skills training scenarios to help clinical trainees (e.g., medical students, residents, nursing students, pharmacy students, etc.). Currently, our scenarios are scripted through information obtained from research team discussions, stakeholder interviews, and literature reviews without ChatGPT assistance. We will then use ChatGPT to prototype scenarios before we program them using simulation software. During prototyping, trainees will practice conversations with a simulated patient role played by ChatGPT. The objectives will be to develop at least one training scenario and prototype it using ChatGPT. The outcome will be a new development process, which we anticipate to be successful and plan to publish in a peer-reviewed scientific journal.

Duration

2/20/25 - 11/20/25

Project Lead and Team Members

Warren Scott Comulada - Director of the Semel Institute Center for Community Health
W. S. Comulada, Dan Weisman, Johnny Hossell

Executive Summary

Enrollment Management seeks to use AI to develop and power an online module to assist students in exploring UCLA majors and minors based on their submitted interests, academic strengths, and personal goals. About 1,500 students are “undeclared” at any given time, and many more change their majors while at UCLA. Additionally, Undergraduate Admission also receives over 173,000 First-Year and Transfer applications from students worldwide. An online module developed by UCLA to provide support for academic exploration serves as a valuable and far-reaching tool to spark interests and provide guidance for our current and prospective students. The project requires the need to map out UCLA’s complete offering of majors and minors and lean on AI to suggest majors, academic pathways and careers based on input data about personal strengths, intellectual curiosity, and desired career outcomes.

Project Goals

This project aims to leverage AI to assist incoming and continuing students to explore and identify potential majors and careers based on input from a series of dynamic intake questions.

The project will utilize AI to define paths and interests associated with UCLA’s 140+ majors and 90+ minors. The project outcome is an integrated module that can offer exploration on “best-fit” majors and careers based on student input. This project will allow campus to service more students by offering a robust exploration for students through a self-guided system.

The project requires guidance and leadership from UCLA IT personnel to leverage AI in developing the online module.

Duration

3/03/25 - 6/02/25

Project Lead and Team Members

Manyee Lieu - Dir Comm Enrollment Management
Gary Clark - AVC Enrollment Management
Ffiona Rees - Exec Dir Undergraduate Admission
Karly Brocker - Senior Assoc Dir Admission
Scott Carter - Deputy Director Enrollment Management

Executive Summary

The use of Generative AI for this project has two major outcomes. On the one hand, we will train AI to gain proficiency in Basque. This language, although it is used in generative AIs we have tried, needs to be trained for basic/colloquial level and also to provide a learning curve where AI should not go beyond—and sometimes should stay a bit behind—student’s own level. This might be an imperfect process at this stage, but something that is critical in order to the VR environment to be productive for the students. On the other hand, AI will provide of generative interactive content as students explore the VR environment, both linguistically and culturally, enriching the VR scenario itself. This can be achieved by different methods (generative text, augmented reality, interpretation of images, etc.). By the combination of a Generative AI in a virtual environment, we do not only achieve the goal of offering a practical and productive—in the interpretative and communicative goals—environment, but we shift the interactive capabilities of the VR given world, which is one of the major goals of any language, namely, interact with a given reality

Project Goals

OpenAI function in this project is to complement the VR sessions we will conduct in Spring Quarter 2025. OpenAI shall deliver the part where students interact with a quasi-avatar entity, in the figure of a narrator of a story, a character that provides clues to explore that VR environment or other possible scenario where the students need to make progress in a task or project by the use of language. These tasks might involve a game-like process, an adventure, or support—correction, assessment and co-evaluation. In the process, OpenAI might be able to provide also cultural content, generate explanations or context that simultaneously might help in the achievement of the project. Finally, OpenAI might be a good source for assessment of the project itself, since we should be able to allocate any data we are generating to provide an interpretative and analytical review of the project, students’ performance and environment’s productivity. The following areas and/or disciplines should benefit from this project: Basque Studies, Basque Language Teaching and Learning, Technology for Language Acquisition, AI Based Learning.

The project objectives:

General Goal: Provide a fully interactive AI assisted VR environment for students to practice and learn Basque language and have an immersive cultural experience.

Language Interpretative Goals: Increase the level of engagement and performance in students by providing a set of generative scenarios where AI can act as a conversation partner.

Communicative Goals: not only with AI but we can design activities where students perform collaborative tasks with the use of generative AI in Basque. This can be further achieved with the newly updated Voice feature in OpenAI

Cultural Goals: set historical or culturally relevant scenarios students can visit in VR format, making it accessible, but also gain agency since they will be in a position to make decisions on how to proceed and progress in these tasks.

The outcomes and results that this project will accomplish if successful.

Make a valid case of generative AI to be useful with the learning small/minor/less spoken languages:
Make a valid case for generative AI be used in language acquisition in combination of cultural content
Make a valid case to expand the use of generative AI and VR environments for a larger project for Basque Language and Culture that would consider the creation of a platform to develop all the potential in this direction.

Duration

1/06/25 - 6/02/25

Project Lead and Team Members

Iker Arranz Otaegui, Ph.D.; Adjunct Assistant Professor in Basque Studies
Jordan Galcyinski
Wendy Perla
Nic Smith
Saad Shaukat

Executive Summary

This proposal outlines the development of a multi-modal, lifelong artificial intelligence (AI) clinical agent system designed to collaborate seamlessly with OpenAI’s advanced large-scale pre-trained models. The system will engage in tasks such as diagnostics, patient communication, and medical research through conversational interactions.

Leveraging publicly available datasets, including MIMIC-III and MIMIC-IV for textual clinical understanding, and PMC-OA, PMC-VQA, and MedMD for multi-modal clinical understanding, the system will be capable of processing diverse clinical data sources. These include electronic health records (EHRs), medical imaging, and patient-reported outcomes, offering a comprehensive, longitudinal view of patient health.
The agent system is designed to interpret complex medical data, assist in the early detection of potential health issues, and deliver actionable insights to healthcare providers. By employing predictive analytics, it will forecast patient outcomes and generate personalized treatment recommendations. OpenAI licenses will enable advanced functionalities, such as medical text analysis, question answering, and automated report generation, while adhering to strict privacy and ethical standards.

Ultimately, this project aims to empower healthcare professionals by providing optimized, data-driven support, improving patient outcomes, and reducing healthcare costs.

Project Goals

The scope of the envisioned experiment focuses on developing and testing a multi-modal, lifelong AI clinical agents system that leverages the public datasets to enhance the quality of patient care through AI-driven insights. The experiment will involve the integration of multi-modal data sources, including patient demographics, clinical notes, laboratory results, medical imaging, and patient-reported outcomes.
This scope aligns with Activate AI Research & Product Development, which develops new code and algorithms to prototype novel AI applications and conduct research on AI use in clinical diagnosis fields.

The goal is to create a system capable of long-term monitoring of patient health, identifying early signs of deterioration, and generating personalized recommendations to healthcare providers.

We aim to publish our work in publicly accessible journals catering to audiences in clinical, medical, and machine learning fields, such as Nature Medicine, Nature Communications, and Nature Machine Intelligence. Additionally, top-tier machine learning conferences like NeurIPS, ICML, and ICLR are strong candidates for dissemination. Besides, we plan to open-source our code on GitHub to get more than 200 GitHub stars.

Duration

1/01/25 - 5/31/25

Project Lead and Team Members

Wei Wang, Leonard Kleinrock Chair Professor in Computer Science University of California, Los Angeles
Haixin Wang, PhD student in computer science, UCLA
Kaiqiao Han, PhD student in computer science, UCLA
Han ZHang, PhD student in computer science, UCLA

Executive Summary

Our project aims to leverage OpenAI's GPT-4 multi-modal technology to streamline and improve remote blood pressure monitoring by capturing handwritten patient blood pressure logs into digital format for integration into UCLA Health's electronic health records (EHR). Hypertension control is a UC-wide and UCLA Health priority, with a steering committee supporting numerous individual projects to improve the quality of blood pressure control. Using AI-enhanced OCR, the system will process and digitize entire pages of handwritten blood pressure data, accurately converting them into structured, electronic data ready for secure transmission via SMART on FHIR. Once transmitted, these digitized vitals will automatically update within the patient’s EHR, ensuring clinicians and population health programs have timely access to remote patient data. By enabling the use of handwritten logs instead of high-tech monitoring devices, this solution aims to reduce costs and improve accessibility for patients, making remote monitoring feasible for broader populations to reduce disparities. This approach supports both the UC and UCLA Health’s goals to improve blood pressure control and patient outcomes through innovative, cost effective, patient-friendly digital health solutions.

Project Goals

Scope and Use of OpenAI Licenses

This project will employ OpenAI's GPT-4o model to recognize and interpret handwritten patient blood pressure logs, digitizing the data to integrate with UCLA Health's EHR system via SMART on FHIR protocols. The OpenAI licenses allow for using GPT-4o’s multi-modal processing capabilities to ensure accurate recognition and structured conversion of handwritten blood pressures into a digitized format that is appropriately categorized into discrete variables including but not limited to date, time, systolic blood pressure, diastolic blood pressure and heart rate.

Objectives

The primary objectives are to:

Accurately transcribe handwritten vital signs data into digital format aligned with SMART on FHIR via GPT’s 4o multimodality model
Compare estimated cost of AI-enabled solution vs. current standard-of-care solution with Bluetooth or cellularly-connected Blood Pressure machines

Outcomes

If successful, the project will create a reliable and cost-effective solution for remote blood pressure monitoring without relying on advanced technology devices, as many patients already have access to a smartphone with camera and non-bluetooth or cellular-connected and integrated BP machine. This will also potentially decrease labor costs currently associated with technology support for the initial set-up and maintenance of these expensive technological solutions.

Duration

12/01/24 - 6/30/25

Project Lead and Team Members

David Cho, MD, Cardiologist
Brian Le, MPH, CPHQ, Program Manager
Charlene Huang, Electrical Engineering c/o 2027
Clyde Villacrusis, Computer Science c/o 2026
Diego Valencia, Computer Science and Linguistics c/o 2027
Alicia Yu, Computer Science c/o 2026

Executive Summary

We are aiming to advance research on leveraging Large Language Models (LLMs) to enhance the decision-making capabilities of Autonomous Driving Systems (ADS). We focus on enabling ADS to interpret, retrieve, and apply traffic regulations dynamically across diverse scenarios. To foster trust and accountability, the project aims to develop a novel, authoritative evaluation framework inspired by DMV driving tests and insurance company methodologies. This comprehensive approach ensures that ADS can demonstrate compliance with legal requirements and safe driving behavior, ultimately enhancing public trust in autonomous driving technology.

Project Goals

Objectives

Traffic Rule Comprehension: Explore LLMs' ability to understand traffic rules and demonstrate this understanding through natural language-based tasks, such as Q&A and scenario evaluations.

Integration with ADS Planning: Investigate how LLMs can assist ADS planning through high-level instructions, parameter tuning, and trajectory waypoint generation.

Development of an Authoritative Evaluation Framework: Design an evaluation system inspired by DMV driving tests or insurance company practices to assess ADS compliance with traffic laws and safe behavior.

Scalability and Adaptability: Ensure that the framework can dynamically retrieve and apply regulations across different states and countries, reflecting diverse legal landscapes.

Duration

1/01/25 - 5/31/25

Project Lead and Team Members

Jiaqi Ma, Associate Professor

Executive Summary

This proposal aims to significantly enhance institutional effectiveness by integrating ChatGPT Enterprise into UCLA’s Endpoint Solutions operations. With a large volume of technical issues across thousands of endpoints, leveraging ChatGPT will allow the Endpoint Solutions team to streamline troubleshooting, improve the accuracy and speed of issue resolution, and deliver a higher standard of customer service. By automating routine processes and providing intelligent assistance for complex technical issues, the project will directly contribute to operational efficiency, improve service delivery, and allow the university to allocate its IT resources more strategically.

Project Goals

The project will focus on using ChatGPT Enterprise to enhance the effectiveness and efficiency of Digital & Technology Solutions' Endpoint Solutions team. The ChatGPT licenses will assist in rapid troubleshooting, reducing time spent on routine issues and allowing more complex cases to be handled with greater precision.

Objectives

Boost institutional effectiveness by optimizing Endpoint Solutions workflows
Achieve faster resolution of technical problems, improving overall operational efficiency
Enhance customer satisfaction and minimize downtime for critical university operations

Duration

11/15/24 - 5/31/25

Project Lead and Team Members

William Dubin, Supervisor, Endpoint Solutions
David Arizmendi – Supervisor, Dept. Endpoint Solutions (Bespoke Operations)
Lena Chen – Sr. Endpoint Support Analyst
Brian Ear – Sr. Endpoint Support Analyst
Aiden Earle – Endpoint Support Analyst
Pulkit Gupta – Sr. Endpoint Support Analyst
Reid Iwane – Supervisor, Endpoint Solutions
Steve Jeong – Sr. Endpoint Support Analyst
Todd Kallemeyn – Endpoint Support Analyst
Clark Kringel – Sr. Endpoint Support Analyst
Mario Medeles – Endpoint Support Analyst
Raymond Oreiro – Endpoint Support Analyst
Oscar Partida – Sr. Endpoint Support Analyst
Ace Selters – Lead Student Endpoint Support Analyst
Perry Usman – Endpoint Enginer
Alex Wansing – Sr. Endpoint Support Analyst
Todd Weber – Director, Endpoint Solutions
Robert Wilkinson – Sr. Endpoint Support Analyst
Jimmy Wu – Endpoint Support Analyst
Gilbert Zirrakyan – Sr. Endpoint Engineer

Executive Summary

Online education has fallen short of its promise to make education universally accessible. Primarily because ineffective learning experiences make it difficult for students to succeed in their studies. The pragmatic reason for ineffective online courses is that faculty and Instructional Designers spend the bulk of their time creating core content, regularly running out of time to include what really matters to students: The faculty’s expertise and experience, the most current research or the most intriguing insights, as well as engaging learning activities for active learning, and individualized tutoring.

With four major US universities, we are engaging in a joint research project to establish a process and prompting protocol by which instructional designers and faculty can create above average quality content for online courses. This includes the automation of core content creation to gain time to create new, interactive learning elements and activities. The objective is to establish a novel framework that can serve as a national standard for the use of Generative AI in online course development. In our current assessment, GPT4 is the only model capable of producing above average quality course content across various disciplines, and an API access would enable us to build interactive, discussion-based learning activities.

Project Goals

Creation of a complete, 4 credit online course using GPT4, including all core content such as lecture texts, slide decks with scripts, assignments with rubrics, and assessments.
Use GPT4 API to create interactive learning elements, personalized learning objects and interactive learning activities in collaboration with faculty.
In collaboration with instructional design teams and faculty from GeorgiaTech, Purdue Online, University of Arizona and University of Illinois Urbana
Champaign, develop a common framework for online course content development with Generative AI, including a prompting protocol and content generation process.
In collaboration with the partner universities, formalize a first standard for the responsible use of Generative AI for course design and development. The process and protocols and standard framework, would empower faculty and instructional design teams across the US, and potentially the globe, to achieve highest quality courses that better engage students and make learning significantly more effective and efficient.

Duration

1/06/25 - 6/30/25

Project Lead and Team Members

Frederick T. Wehrle, Assistant Dean for Academic Innovation and Learning

Executive Summary

Laboratory safety is of paramount importance in chemistry research and teaching laboratories and a safe workplace is the responsibility of the principal investigator (PI) and university. Safe practices in the laboratory are enforced through lab meetings, online trainings, hands-on practice, and frequent feedback from the PI. While accidents and near misses can happen at any point, novice researchers may not be keenly aware of specific safety hazards, how to identify proper standard operating procedures, or safe experimental design. We propose to use prompt engineering and domain-specific resources to guide ChatGPT in assisting novice lab researchers in the preparation of experimental safety plans (ESP). Prompt engineering will enable ChatGPT to construct an ESP informed from diverse sources with an undergraduate or novice graduate student starting from a proposed synthetic protocol. After formulation, researchers can engage with the platform in a question-and-answer format prior to evaluation of the ESP by the mentor.

UCLA EH&S has nominally agreed to assist in the evaluation of LAB SAFE. While LAB SAFE does not replace feedback from a supervisor, the platform will greatly enhance students’ education and training on laboratory safety, while multiplying the limited resources of EH&S.

Project Goals

This proposal aims to develop a lab platform that is capable of providing personalized assistance to undergraduate and novice graduate lab researchers in the design and conduction of lab experiments in a safe manner. The specific aims of the proposal are (I) platform and repository development (II) consultation and validation of platform with an EH&S representative (III) Implementation in an active chemistry research lab and undergraduate teaching laboratory (IV) qualitative and numerical comparison with standard lab practices.

ChatGPT recently implemented a “Lab Safety Advisor” This project has great potential to introduce students in lab courses to common safety practices, however, UCLA and lab specific guidelines are not considered without considerable feedback in the “Lab Safety Advisor”. We aim to address this capability gap and develop a UCLA specific platform which supports the role of PIs and EH&S in laboratories.

Duration

1/15/25 - 6/01/25

Project Lead and Team Members

Matthew Nava, Assistant Professor, Department of Chemistry and Biochemistry
Anneke Talke - UCLA ugrad class of 2026
Kevin Liu - UCLA grad student 2027
Lina Zarnitsa - UCLA grad student 2027
Vito Lin - UCLA grad student 2027

Executive Summary

Project Goals

Duration

1/15/25 - 8/01/25

Project Lead and Team Members

Warren Scott Comulada, Professor-in-Residence
Dallas Swendeman, Professor-in-Residence
Yue Ming Huang, Adjunct Professor
Johnny Hossell, Information Systems Analyst 2
Dan Weisman, Learning Experience Designer
Kelsey Ishimoto, Doctoral Student, Department of Biostatistics

Executive Summary

This project aims to leverage OpenAI's enterprise licenses to improve and automate the generation of QA test cases directly from Jira epics and stories. By using advanced natural language processing (NLP) models, we will transform Jira stories into detailed QA tests, ensuring comprehensive coverage, consistency, and efficiency in the QA process. The project will focus on optimizing test creation workflows to reduce manual effort, enhance accuracy, and align closely with development cycles in agile methodologies. This initiative will significantly reduce the time spent on test case creation while increasing overall test coverage and quality.

Project Goals

This project will explore how OpenAI licenses can be used to automatically generate QA test cases by analyzing Jira epics and stories. Our objective is to improve the speed, consistency, and quality of test coverage in agile environments. By automating test creation, we aim to reduce human error and ensure that tests align more closely with evolving development requirements. Success will be measured through improved test case accuracy, reduced manual QA efforts, and enhanced coverage. The project will also aim to integrate this solution into existing Jira workflows for continuous deployment.

Duration

11/06/24 - 6/13/25

Project Lead and Team Members

Jake Cabrera, DevOps Engineer
Inessa Tadevosyan, Senior QA Analyst

Executive Summary

This proposal seeks to secure ChatGPT Enterprise licenses and integrate with Lucidchart to accelerate process mapping and process improvement activities. The primary focus is to enhance our ability to efficiently gather, analyze, and visualize business requirements, improving both the speed and quality of these efforts. ChatGPT will assist in digesting stakeholder requirements and creating first-pass process maps, while Lucidchart will visualize these maps for better clarity and collaboration. This will streamline CRM development work in areas like data integration, user experience enhancement, and system automation. This integration will reduce the manual effort required in translating stakeholder needs into actionable process designs.

Project Goals

The project will focus on using ChatGPT Enterprise licenses integrated with Lucidchart to streamline the process of mapping and improving CRM-related workflows. The primary objective is to reduce the time and effort spent in process mapping while increasing the accuracy of stakeholder requirement translation into actionable system configurations. A secondary objective is related to change management and emerging requirements, ensuring the process mapping stays accurate through-out the entire lifecycle of the project. By automating parts of the requirement-gathering process, we aim to accelerate our CRM development timelines and reduce bottlenecks. Successful outcomes will include the creation of efficient process maps, more accurate system requirements, and faster implementation times for CRM enhancements and updates.

Duration

11/18/24 - 4/30/25

Project Lead and Team Members

Zach Przybilla, Director CRM Products
Tina Gonzales, CRM Product Owner
Shira Stanley, Salesforce BSA
Sergio Zambrano, ServiceNow BSA

Executive Summary

To remain informed about progress and keep pace with advancements in AI, this project proposes the development of an AI-powered platform that integrates the UCLA Career Guide and career development and education specifically developed for UCLA students, with a user-friendly interface. This platform aims to provide personalized career guidance, empowering students to explore career pathways beyond higher education, receive tailored advice on resumes and cover letters, and prepare for internships, jobs and graduate school interviews. By offering real-time access to UCLA Career Center resources, the platform will enhance student support, streamline career services, and promote the innovative use of AI in education.

Creating a GPT for career resources will ensure accessibility for all students, campus partners and faculty, fostering collaboration and community across campus departments. Planned training through our current I-CAN program (which educates campus partners on career practices and resources) will further build capacity in AI technologies. The goal is to continuously expand this model with vetted resources, providing UCLA Bruins, staff and faculty with the best tools to enhance and connect students’ academic potential to their individual career paths.

Project Goals

This project aims to develop an AI-powered career guidance platform using OpenAI to integrate the UCLA Career Guide with resources like industry reports and job market data. The platform will provide personalized, on-demand career advice, improving the accessibility and equity of career services for Bruins.

Objectives

Offer Initial Guidance: Support students exploring career options by providing direction and clarifying key questions.
Advance Career Readiness and Accessibility: Equip students with personalized education, tools and resources to succeed in their career.
Familiarize with AI Technologies: Introduce UCLA’s community to generative AI through a platform dedicated to Bruin success.

The project will also equip faculty and staff with emerging industry insights, enriching their ability to guide students. If successful, it will increase student engagement (i.e. UCLAOne, FDS), enhance career outcomes (resumes, exploration), and inform staff and faculty on effective AI use, fostering collaboration aligning UCLA’s academic and career goals.

Duration

1/08/25 - 6/01/25

Project Lead and Team Members

Andre Philippo - Assistant Director, Undergraduate Career Education & Development

Project Managers

Jessica Oviedo - Senior Assistant Director, Undergraduate Career Education & Development

Maria José (MJ) Hidalgo Flores - Assistant Director, Undergraduate Career Education & Development

Luke Cheves - Industry Relations Manager, Industry Relations & Experiential Learning

Quality Assurance Team

Armine Kulikyan - Assistant Director, Undergraduate Career Education & Development
Jasmine Miranda - Program Specialist, Undergraduate Career Education & Development
Melisa Garcia - Assistant Director, Undergraduate Career Education & Development
Roni Lavi - Assistant Director, Undergraduate Career Education & Development
Sanaz Nabati - Assistant Director, Undergraduate Career Education & Development
Timothy Mar - Industry Relations Manager, Industry Relations & Experiential Learning
Arielle Cunanan - Career Peer Intern, 4th year, Psychology
Vyvy Nguyen - Career Peer Intern, 4th year, Biology

Executive Summary

Imagine an AI system capable of unlocking the full potential of athletes by analyzing every movement, providing real-time corrections, and driving human performance to new heights. We propose a cutting-edge research project to develop and validate such a system, leveraging advanced AI technologies, including OpenAI’s reasoning models and state-of-the-art pose estimation algorithms. The system will analyze human motion to deliver actionable insights, optimizing athletic techniques and enhancing performance at professional levels.

Our approach includes developing a prototype capable of precise analysis of athletic form and validating its effectiveness through small-scale trials with athletes. By harnessing advanced AI reasoning, the system will provide robust optimization of techniques, offering insights that redefine training for elite athletes. This initiative aims to debut with UCLA sports teams, demonstrating its transformative potential in real-world applications.
By laying the foundation for future innovations, this project also positions Los Angeles as a leader in sports technology, with the potential for broader application at the 2028 Olympic Games. With precise, data-driven feedback and optimization, this AI system promises to set a new standard for athletic training and performance enhancement.

Project Goals

This project will leverage OpenAI licenses to develop an AI-powered system that uses computer vision and pose estimation to provide real-time, automated feedback on athletic movements. The scope includes training AI models to analyze human motion and offering actionable insights to optimize athletic performance. The focus will be on refining pose estimation algorithms and integrating them with OpenAI’s reasoning capabilities to deliver precise, personalized feedback.

Objectives of this project are:

Design and develop a prototype system that utilizes AI to analyze and assess human athletic form with precision.
Validate the system’s impact on enhancing individual athletic performance through targeted, small-scale trials with athletes.
Create a robust framework for optimizing techniques, leveraging OpenAI’s advanced reasoning models to generate actionable insights for professional-level athletic performance.
Lay the groundwork for future applications, positioning the system to be a key innovation in the 2028 Olympics.

If successful, the project will produce a functional AI system capable of enhancing athletic performance, advancing sports science research, and establishing UCLA as a leader in AI-driven human performance optimization.

Duration

1/01/25 - 6/01/25

Project Lead and Team Members

Jonathan Ouyang - UCLA Class of 2028
Yuchen Cui - Professor of Computer Science UCLA
William Jiang - UCLA Class of 2028
Daniel Wu - UCLA Class of 2028

Executive Summary

BruinAgent is an innovative, accessible next level solution for facilitating stronger communication and engagement with UCLA researchers. A powerful, valuable AI assistant for helping users connect with the most fitting Office of Advanced Research Computing (OARC) experts quickly and easily, BruinAgent is a customizable, scalable, and reusable product. During this phase of our project, our team will complete the development and testing we started as part of OARC’s AI Accelerator program so that we can deploy our AI assistant on the oarc.ucla.edu website. We will amplify performance, enhance server security, expand the knowledge base, incorporate user feedback, and conduct QA testing. This will strengthen the ability to offer an intuitive, guided experience rooted in the principles of personalized assistance. BruinAgent carries the unique potential to expand as it becomes further enhanced and to make a strong, positive impact into the future.

Project Goals

Our objective is to deploy BruinAgent to the OARC website, initially as a private alpha release, and ultimately as a production release. BruinAgent will perform the action of guiding users through the process of contacting specific experts at OARC, presenting custom options and explaining services based on user inputs. We expect an increase in the number of users successfully contacting OARC experts, and we also will provide a further enhanced user experience.

Duration

12/18/24 - 6/01/25

Project Lead and Team Members

Lauren Cullen - Project Lead and Product Designer
Anthony Doolan - Technical Lead and Software/Hardware Engineer
Qiyang Hu - AI Solutions Architect
Catherine Yanga - User Testing/QA Analyst and Prompt Engineer

Executive Summary

This project aims to develop a web-based platform that utilizes OpenAI licenses to provide streamlined, step-by-step troubleshooting resources for engineers and technicians. The platform will feature instructions derived from existing ServiceNow tickets, addressing various network issues such as access point configuration, DNS settings, and Layer 1 infrastructure problems. By referencing ServiceNow links, the project will showcase proven solutions to motivate engineers and enhance their problem-solving capabilities. This initiative will improve operational efficiency and foster a culture of knowledge-sharing, allowing users to access valuable resources anytime, anywhere.

Project Goals

The scope of this experiment involves creating a web platform that consolidates information from ServiceNow tickets and other available resources. OpenAI licenses will be used to generate and structure content, ensuring that instructions are clear and accessible. Project objectives include providing easy access to troubleshooting steps, improving response times for network issues, and creating a repository of knowledge for future reference. Successful outcomes will consist of reduced ticket resolution times and increased confidence among users in addressing technical challenges.

Duration

12/08/24 - 5/05/25

Project Lead and Team Members

Eren Erener, Network Engineer, Wireless/Radio

Executive Summary

This project focuses on developing AI agents designed to improve student understanding of AI models, guide them in identifying the best scenarios for automation with APIs, and support faculty in embracing AI technologies confidently. The initiative provides interactive tools for students to experiment with foundational AI concepts and automation, clarifies when AI solutions are appropriate, and ensures equitable access for both students and faculty. The ultimate aim is to enhance practical AI skills and encourage a positive, informed approach to AI integration within educational settings.

Project Goals

The project's main goal is to create AI agents that help students understand AI models, recognize when an AI solution is appropriate, identify optimal automation scenarios using APIs, and promote faculty optimism about AI integration. Specifically, the project aims to:

Develop custom AI agents for students to explore and experiment with foundational AI models.
Illustrate practical use cases for API-driven automation and clarify situations when AI solutions are unnecessary.
Enhance student skills in evaluating and applying AI and automation effectively.
Provide equitable, hands-on access for students and faculty to deploy custom AI agents in real-world applications.
Encourage positive faculty attitudes towards adopting AI technologies in teaching.

Duration

12/19/24 - 6/30/25

Project Lead and Team Members

Tina Austin, Lecturer and AI adoption specialist, UEI

Executive Summary

Significant strides in artificial intelligence over the past few years have enabled robotic systems to interpret and interact with the world in increasingly versatile ways. The large, often multi-modal, datasets that are used to train modern learning-based systems endow robots with capabilities like scene understanding and commonsense reasoning. However, the safe integration and reliability of these learned models, like large language models (LLMs) and vision language models (VLMs), for robotics applications still remains an open problem. Learned perception systems fail to identify objects correctly and LLM-based planners hallucinate their outputs leading to unsafe robot behavior downstream. The need for reliable outputs is especially critical for robots navigating in real-world environments that are dynamic and unpredictable where safety is of utmost importance.

Consider a disaster-struck region wherein the robot needs to be able to correctly identify victims needing help versus rubble that the robot needs to avoid to maintain safety. Foundation models like GPT-4o by OpenAI are able to reason about the terrain for motion planning and exploration in environments that are safety-critical. However, the model is not always reliable or accurate in its reasoning. This highlights areas where GPT-4o's terrain assessment could be refined using human feedback. We propose to quantify the uncertainty of VLMs for robot motion planning tasks while incorporating human feedback to continuously improve the model output.

Project Goals

We use statistical tools like conformal prediction to quantify the uncertainty of the VLM outputs, like that of GPT4o, during robot motion planning. However, a key limitation of conformal prediction is that its calibration guarantees are typically valid only under conditions where the model and environment remain consistent with the calibration dataset, limiting its applications in new or shifted real-world scenarios. We propose to tackle this challenge directly by computing the distribution shift in the real world. Our approach tackles this limitation by actively monitoring and adapting to distributional shifts observed in real-world applications.

We employ two strategies: first, the VLM-based motion planner's performance is monitored in real time, with human operators providing feedback after each executed plan to confirm successful task completion. Second, if the operator identifies additional insights or performance preferences, this feedback is utilized to finetune GPT4o accordingly, by improving its prompt. Recognizing that standard prediction sets may lose validity post-finetuning or in out-of-distribution contexts, we propose the construction of distributionally robust prediction sets, validated through continuous human feedback, to ensure reliability even as the model evolves. These theoretical guarantees will be validated in new types of environments not limited to indoor, home environments but also rugged terrain for search and rescue applications. Our existing work on risk-aware traversability evaluation and planning can be greatly improved with semantic reasoning of the environments. We will make our autonomy pipeline publicly available for use.

Duration

2/01/25 - 9/01/25

Project Lead and Team Members

Anushri Dixit, Assistant Professor
Qizhao Chen, PhD Student, Class of 2029, UCLA MAE
Shoh Nishino, Undergraduate Student, Class of 2026, UCLA MAE

Executive Summary

This proposal aims to harness the power of Generative AI to advance wildfire management and smart city planning. Building on OpalAI's successful NASA SBIR project, FireGPT, and DOT SBIR project, SMART Tools, we propose developing a unified AI framework that integrates spatial intelligence and generative AI models. This framework will enhance decision-making in wildfire risk assessment and urban infrastructure planning. By leveraging NASA's Earth Science data and DOT's urban datasets, the project will create real-time, actionable insights for city planners and emergency responders. The initiative aligns with UCLA's strategic goals by fostering interdisciplinary AI research and developing innovative solutions with significant societal impact.

Project Goals

The proposed project aims to develop an innovative AI system by integrating OpalAI's successful initiatives in both wildfire management (FireGPT) and smart city planning (SMART Tools). Utilizing OpenAI licenses, we will create a unified AI platform that leverages spatial intelligence and generative AI models to enhance decision-making for urban planning and wildfire risk assessment. The experimental scope includes developing models that process large-scale spatial data for real-time insights.

Project Objectives

Develop a generative AI framework that integrates data from NASA and DOT projects to provide enhanced decision-support in urban and wildfire management.
Utilize OpenAI’s capabilities to refine AI models capable of generating predictive analytics from diverse datasets.
Create a decision-support tool that offers real-time insights for city planners and emergency responders.

Outcomes and Results

The project will result in a robust AI platform providing accurate wildfire risk assessments and optimized urban planning solutions. This will lead to improved public safety, efficient resource allocation, and enhanced infrastructure planning. The integration of advanced AI techniques will position OpalAI as a leader in spatial intelligence-driven solutions, offering scalable applications for broader geographic and sectoral expansion.

Duration

1/01/25 - 7/31/25

Project Lead and Team Members

Dr. Ryan Alimo, MBA Student, Anderson

Executive Summary

UCLA’s Office of Advanced Research computing plans to explore the use of GEN-AI in remediating documents created in environments typically utilized in the Higher Education work to be compliant with WCAG 2.1 Accessibility standards. We plan to have staff and student accessibility engineers research effectiveness in working with ChatGPT AI LLM to remediate documents created in these environments: Microsoft Word, Google Docs, Adobe’s PDFs, Google Slides, Microsoft Powerpoint, Google Sheets and Microsoft Excel. We will test the remediated documents against other campus remediation tools, potentially including but not limited to the Bruin Learn ALLY tool if possible. If we learn that these prompts can be developed to do this remediation successfully, we will build out a password protected DCP web-based toolkit that leverages the Enterprise API by using tokens and create a an easy to use prototype of a remediation web-based toolkit.

Project Goals

The scope of this project is to research what Open AI's ChatGPT LLM is capable of in the way of WCAG 2.1 remediation of documents typically utilized in the Higher Ed work environment. Goals include learning if this is an effective use of LLMs, and if yes, whether a prototype to re-use the successfully researched prompts can be turned into a web based service or product.

Duration

1/06/25 - 6/01/25

Project Lead and Team Members

Rose Rocchio, Director of Mobile Web Research & Accessibility
Davida Johnson - Executive Director of the Office of Advanced Research ComputingTravis Lee - UCLA DCP Coordinator
Chris Patterson - Senior Accessibility Engineer & Web Developer
Carolanne Link - UWAI Program Manager
Sal Santa Ana - DCP Testing Lead
Eleanor Koehl - Senior Program Manager for Research Facilitation
Nate Jacobs - AI Project Manager Advisor
Melissa Chang - Student Accessibility Engineer
Nicholas Shinghal - Student Accessibility Engineer

Executive Summary

This project proposes the development of an AI-powered chatbot that will provide support to students across all Student Affairs and College/Undergraduate Education websites at UCLA. By leveraging OpenAI’s large language models (LLM), the chatbot will be trained on publicly available data from UCLA’s official websites, offering students quick and consistent answers to their questions. This solution is intended to assist front-end support staff by automating responses to common queries, thereby reducing their workload and allowing them to focus on providing personalized support to students who need it most. The current chatbots employed by individual departments are limited by their reliance on custom FAQs, which require frequent updates and have proven ineffective in resolving queries. An LLM-based system will offer a more dynamic, intelligent response mechanism, capable of answering differently framed questions with the same core information. This approach promises to improve service quality, reduce redundancy, and standardize the student support experience across UCLA's departments.

Project Goals

The scope of this project involves designing, training, and deploying an AI chatbot that will assist UCLA’s Student Affairs and College/Undergraduate Education departments. The chatbot will use publicly available data from UCLA websites to answer student queries, thus reducing the workload on front-line support staff. The primary goal is to ensure that students have access to accurate, timely, and consistent information, no matter which department or website they visit. By training the chatbot on existing web content and deploying it across multiple platforms, the project will ensure seamless support for students and a significant reduction in the administrative burden on UCLA’s staff.

Duration

1/03/25 - 6/01/25

Project Lead and Team Members

Arun Pasricha (CIO, Student Affairs), Christian Spreitzer (Director, Undergrad IT)
Praveen Dugar
Nathan Lai
Donny Morada
Matthew Geddert
John Hirning
Seng Chea

Executive Summary

The “Bruin Workplace IT Bot” project proposes an innovative solution to streamline IT support and operational analysis for our Workplace IT products. This solution will leverage an OpenAI-powered chatbot to enable natural language interaction and analysis of real-time application and server log information, along with product source code. By leveraging existing data sources like AppDynamics and Splunk, the chatbot will simplify troubleshooting, reduce time-to-insight, and offer a unified interface for querying operational data.

The project prioritizes data security and privacy, ensuring compliance with UCLA’s IT security policies. No sensitive data (P3/P4) will be included in the project scope, and transparency will be maintained in the AI system’s workings. The project team comprises experienced IT professionals with expertise in software development, data administration, IT architecture, and actively learning the AI technologies.

The project is estimated to be completed within 20 weeks, utilizing 10% of team members’ allocated time and voluntary additional hours. There will be no impact on regular business operations.

This project promotes the early adoption of AI technologies within the project team. The chatbot will benefit various audiences, including IT product teams, stakeholders, support staff, and non-technical teams, by enhancing operational efficiency, streamlining troubleshooting, and improving user experiences.

Project Goals

To explore creative uses of generative AI technologies.
Promoting shared learning within the project team and collaborative efforts in leveraging AI technologies in Digital & Technology Solutions products and services.
To equip the team with the skills needed to effectively utilize AI technologies.
To deliver a minimum viable product (MVP) with the above listed features, within 20 weeks of receiving the OpenAI licenses.

Project’s initial scope for the MVP is limited to the following:

Workplace IT Products in scope:

BruinCard
Kronos
TRS
FSPH
ODMP Online Directory
Employee of the Month

Types of data sources in scope:

Application log files
Infrastructure log file
Database log files
APM data from AppDynamics
Select GitHub Repositories
Select Internal Documents

Product Features in scope:

Ability to ingest “pre-selected” real-time data sources
Ability to fine-tune the OpenAI models with our custom data
Ability to interact using natural language to query the operational data

Duration

11/01/24 - 3/28/25

Project Lead and Team Members

Jey Ramakrishnan, Manager, Workplace Products
Krishna Seelam - Senior Software Developer

Executive Summary

To ensure a smooth launch of Bruin Financial Aid, it is crucial to understand and adapt the current process to prevent an influx of urgent screen access requests that could disrupt workflows. This involves documenting and streamlining the screen access request workflow to guarantee timely access to student records, admissions, and financial aid products. Additionally, the escalation process must be clarified to handle urgent requests efficiently without interrupting applicant workflows.

To address this, we are creating a swim lane mapping diagram that will help align stakeholders by showing which department is responsible for each step. As a process improvement initiative, this map will also be used to identify new metric targets and assess which steps can be removed or modified for a new ServiceNow workflow.

The plan is to see how we can make use of AI and build a Q&A tool for staff to use about what type of screen they should request access to. We have so many different screens. It could give them answers about how to request access, what type of screen to request access to, and more based on the documentation we feed it.

Project Goals

Develop an AI-Powered Q&A tool for staff to use.
Provide clear instructions on how to request access and the specific types of screens available.
Integrate with Bruin Financial Aid Processes:
1. Analyze and align the access request process with Bruin Financial Aid to prevent a surge of urgent requests during the tool’s launch.
Improve communication with applicants and DSAs throughout the process to build trust and reduce duplicate inquiries.

Duration

10/11/24 - 9/01/25

Project Lead and Team Members

Krithik Udayashankar, Business Systems Analyst, Admissions and Financial Aid, UCLA Digital & Technology Solutions
Anna Ahearn, Product Manager, Admissions and Financial Aid, UCLA Digital & Technology Solutions.
LaShawn Maddox, Product Manager, Student Records & Engagement, UCLA Digital & Technology Solutions.
Vicki Hou, Manager, BAR Student Products, UCLA Digital & Technology Solutions.
Deepu Kaimal, Supervisor, Financial Aid IT Products, UCLA Digital & Technology Solutions.

Executive Summary

Future clinicians need strong clinical reasoning (CR) skills to critically evaluate and effectively utilize an exponentially increasing number of medical AI tools. Medical trainees have difficulty obtaining the 10,000 hours of deliberate practice needed for achievement of expert level CR skills in traditional medical curriculums.

Twenty four medical student groups (6-8 students/group x 24 groups = 176 students total) will be randomized to the commercially available Diagnosing Wisely Curriculum versus a custom UCLA DGSOM AI Clinical Reasoning Curriculum (UCLA AI CR Tutor) in order to compare outcomes in student CR skills. The UCLA AI CR Tutor is a chatbot powered by ChatGPT 4o and paired with GraphRAG to enhance accuracy and emphasis on clinical reasoning and specialty specific medical literature. The UCLA AI CR Tutor will guide students through faculty designed clinical cases. CR skills will be assessed in the control and experimental groups at baseline, midway, and at the conclusion of the study. As it is not educationally feasible for one student group not to receive any additional CR content, the two aforementioned experimental groups will be compared to the class of 2027 (pre-intervention) via year end OSCE (objective structured clinical examination) scores.

Project Goals

The UCLA AI CR Tutor aims to establish strong clinical reasoning skills for all first year medical students before they see patients in their second year, as measured by improvements in OSCE and R-IDEA scores on standardized cases. This can improve patient care and future physicians’ ability to critically evaluate the output of AI risk stratification and diagnostic tools in medicine.

Alignment Area #2: Activate AI research and Development. To our knowledge, the UCLA AI CR Tutor (with or without GraphRAG) is a completely novel application of generative AI to the field of clinical reasoning in medical education. We aim to show that this first iteration of the UCLA AI CR Tutor is at least non-inferior (for student CR skills outcomes) to a commercially available CR Curriculum, as this will represent a pivotal breakthrough for the field of clinical reasoning in medical education.. This is because the UCLA AI CR Tutor is much more cost effective, and has much greater potential for customization and expandability to fit any level or specialty of medical training than traditional CR curriculums.

Duration

1/28/25 - 6/01/25

Project Lead and Team Members

Serena Wang, MD. Assistant Professor, Chair of UCLA AI in Medical Education Council
Anders Garlid, Phd. Post doc , Medical Imaging and Informatics lab
Sarah Gustafson, MD , Director of Foundations of Practice Course. UCLA David Geffen School of Medicine
Peter Quieos, Director of Foundations of Practice Course. UCLA David Geffen School of Medicine

Executive Summary

The prevalence of burnout in academic medicine is alarmingly high at 45 - 75 percent, partially because time protection for the full scope of teaching activities is difficult. Recent research has demonstrated the utility of LLMs in providing useful assessment and feedback for medical learners’ notes and research manuscripts. This AI Medical Teaching Assistant RCT (AI Med TA RCT) aims utilize Generative AI to help faculty decrease time spent on composing written feedback and increase the frequency and quality of student feedback.

Intervention Group: Faculty will receive prebuilt generative AI chatbots that provide suggestions for written homework and OSCE (Objective structured Clinical Examination) feedback. Faculty will also receive basic Generative AI education such as prompt engineering, ethical, privacy, and bias considerations so that they may be empowered to use LLMs to more effectively prepare and teach in a classroom with students of varying needs and abilities. Concurrently, faculty can model safe and effective use of Gen AI tools in medicine and encourage productive conversations on Gen AI use in studying and research with medical students during small group classes.
Control Group: Faculty are provided with general UCLA AI Teaching resources including UCLA Gen AI Guidance, Copilot, and LinkedIn training.

Project Goals

Decreased time spent on written student feedback
Decreased teaching related burnout
Increased comfort with and use of generative AI in various teaching endeavors in the classroom
Increased amount and /or quality of written feedback on OSCE’s and assignments
Increased engagement in the classroom
Increased safe and effective use of generative AI in the classroom

Duration

1/28/25 - 6/01/25

Project Lead and Team Members

Serena Wang, MD. Assistant Professor, Chair UCLA AI in Medical Education Council
Anders Garlid, PhD., Post Doc, Medical Imaging and Informatics Lab
Sarah Gustafson, MD, Director of Foundations of Practice Course, UCLA David Geffen School of Medicine
Peter Quiros, Director of Foundations of Practice Course, UCLA David Geffen School of Medicine

Executive Summary

The Bruin Learn Center of Excellence, in collaboration with the Teaching and Learning Center, plans to collect qualitative data through user stories from UCLA instructors regarding educational technology tools. This data will come from interviews, focus groups, and conversations with instructors, instructional designers, and local support teams. The focus is on identifying how technology tools address challenges like scalability, supporting synchronous and asynchronous learning, and reducing the reliance on physical spaces.

Using OpenAI, the collected data will be organized and coded to identify key themes, arguments, sentiments, and patterns. The process begins with data cleaning to remove sensitive or protected information, followed by prompt engineering to synthesize the data, uncover major themes, and generate insights from transcripts. Once OpenAI processes the data, the PI and team will validate the findings, organize them, and make recommendations. These may include suggestions for adopting tools, enhancing features, or new requests for vendors. Insights from the data may also highlight emerging areas for further investigation into the effectiveness of educational technology tools. This proposal supports UC President Michael Drake's 2030 agenda to expand opportunity and excellence through educational technology.

Project Goals

Scope

Project scope will include 20 or more interviews with a representative sample of instructors using educational technology tools for instructional delivery, content management, media, collaboration, assessment and grading. The scope will be limited to credit bearing undergraduate instruction at UCLA in the 24/25 academic year.

Objectives

To develop effective practices in use of AI to interpret qualitative data to guide educational tool adoption and enhancement.
To facilitate inductive coding of transcribed interviews from instructional teams to assess current educational landscape and provide recommendations for educational technology tools.
To increase capacity for institutional research on educational technology
To promote AI literacy for teams involved in institutional research

Outcomes

The project proposes to achieve the following outcomes:

Report on educational technology adoption and use for purposes of effective deployment of resources
Report analysis of qualitative data in the form of user stories to gain insight on effective pedagogical practices using educational technology.
Develop procedures to use OpenAI to code and synthesize qualitative data for institutional research.

Duration

12/02/24 - 6/02/25

Project Lead and Team Members

Alan Roper, Supervisor of Educational Technology Tools (Media & Collaboration), Bruin Learn Center of Excellence
Nate McKee, Director, Bruin Learn Center of Excellence
Senna Hanner Zhang, Graduate Teaching and Learning Assistant, Bruin Learn CoE; Master of Library & Information Science (2025)
Matt Heinlein, Supervisor, Canvas LMS & Campus Integration, Bruin Learn CoE
Andrew Jessup, Supervisor, Assessment, Grading & Proctoring Tools, Bruin Learn CoE

Executive Summary

This project proposes developing an AI-driven professional development platform that integrates with the University of California’s Career Tracks job architecture and the systemwide learning management system (LMS). Leveraging OpenAI, the platform will offer personalized career guidance, identify skill gaps, and recommend a wide range of learning resources—including specialized courses from third-party vendors such as Pluralsight, LinkedIn Learning, and Udemy—based on individual career aspirations and job roles. It will streamline career development by generating tailored learning paths and mentorship opportunities for staff, enhancing engagement and alignment with university goals. By using OpenAI licenses, the system will analyze staff roles within Career Tracks, skill profiles, and development needs to create AI-powered recommendations that dynamically adapt as employees progress. The platform’s expected outcomes include increased participation in learning programs, improved staff satisfaction, and better skill alignment with labor market needs.

Project Goals

The project aims to develop and pilot an AI-driven professional development platform for UCLA staff, integrating with the Career Tracks framework and LMS. The platform will offer personalized learning recommendations, identify skill gaps, and facilitate mentorship pairing. OpenAI licenses will be used to power the AI-driven analytics and recommendation engine, which personalizes career development paths based on real-time data.

In addition to personalized career development and AI-powered recommendations, the platform will explore integration with third-party learning platforms such as Pluralsight, LinkedIn Learning, and Udemy. This will provide staff access to a broader range of specialized learning resources, aligned with Career Tracks and tailored to individual career needs.

Project Objectives

Provide tailored career growth paths using generative AI.
Increase staff engagement with learning programs.
Enhance alignment between staff development and organizational needs.

Expected Outcomes

30 percent increase in engagement with learning resources through personalized pathways and tailored recommendations.
20 percent improvement in skill alignment with job requirements, ensuring staff development meets organizational needs.
Increased access to specialized learning resources through third-party platforms, contributing to an additional 10% increase in overall engagement with learning programs.

Duration

1/27/25 - 6/30/25

Project Lead and Team Members

Rashmi Umdekar - Manager, HR Information Technology Products

Executive Summary

Large public schools like UCLA have low counselor to student ratios, leaving many students with little guidance on course planning. Navigating through the plethora of courses offered is overwhelming, especially for students pursuing double majors, switching majors, or planning around specific academic goals. To address this, we are developing Bruinbot, a LLM-backed prototype to help UCLA students effectively plan their academic paths. Simplifying schedule planning elevates the academic experience of students, reduces burdens on academic counselors, and increases institutional efficiency aligning with ITS’s goals. The tool will leverage the courses, classes, and dictionary GET APIs provided by the UCLA Registrar to gain access to public information on courses. Combined with user-provided information (intended major(s), graduation date, breaks, etc.), we will use prompt engineering to feed a GPT-4 model, licensed from OpenAI, with deeper context on course offerings and preferences. No real student data will be stored; mock data will be used for the prototype and directly passed into the query, ensuring security through OpenAI Enterprise’s parameters. To ensure ethical considerations, we will implement checks in our prompt-engineered queries (reference mockups) to ensure Bruinbot is inclusive and directs students to counseling for further queries or if unable to generate data.

Project Goals

We will develop a prototype of an LLM-backed course planner using one OpenAI license, leveraging GPT-4 to generate personalized academic schedules. The tool will consider user-specific information such as their majors, completed courses, desired course load, and summer class plans. By integrating the UCLA Registrar's GET APIs, we will access public course data (e.g., course timings, prerequisites) to guide the LLM's recommendations. The project’s goal is to create an AI-powered academic planner that significantly enhances students' ability to navigate their academic journey, offering tailored schedules aligned with their academic goals. Successful implementation will lead to improved academic performance, reduced counselor workload, and better resource management. Since we are building a prototype, no real student data will be used. Instead, mock user data will be used throughout the development and testing phases to ensure the model generates accurate and personalized schedules based on the user’s input and university requirements.

Duration

1/01/25 - 5/30/25

Project Lead and Team Members

Ananya Anand, 2nd year student, UCLA Henry Samueli School of Engineering (Department of Computer Science)
Clifford Kravit, Program Manager, UCLA Health IT, DGIT
Rohan Sinha, 2nd year student, UCLA Henry Samueli School of Engineering (Department of Computer Science)

Executive Summary

Customizing engineering programs at UCLA Extension presents both opportunities and challenges. To enhance course curriculum development and better serve technical managers, the ECP-TMP department is exploring innovative solutions, including AI-driven insights. By analyzing enrollment trends, student feedback, and demand data, the team aims to refine course offerings and improve educational outcomes. A trial utilizing OpenAI is currently underway to assess its potential in supporting data-driven decision-making. This initiative will help determine whether adopting AI tools aligns with UCLA Extension’s mission to provide high-quality continuing education.

Project Goals

UCLA Extension’s Technical Management Program (TMP), a 40-hour leadership program for technical managers from federal agencies and corporations, is exploring innovative ways to enhance course offerings. In March 2025, AI-driven analysis will assist in evaluating student feedback, optimizing course selections, and identifying opportunities for program improvement.

By leveraging data insights, the program aims to continuously enhance the learning experience and expand its impact. The goal is to refine curriculum choices based on evolving industry needs and further elevate the program’s strong reputation. Learn more about TMP here: https://www.uclaextension.edu/engineering/technical-management-program.

Duration

9/15/24 - 12/31/25

Project Lead and Team Members

Nicole Lim, Program Representative, UCLA Extension
Joon Lee, Program Director, UCLA Extension
Gina Springer, Program Manager, UCLA Extension
Alex Alonzo, Senior Systems Engineer, UCLA
Anna Ahearn, Product Manager, UCLA

Executive Summary

We introduce Reinforcement Learning with Specialist Feedback (RLSF), a framework extending the Reinforcement Learning with Human Feedback (RLHF) paradigm to integrate domain-specific expertise into generative AI models. RLSF enables the creation of digital twins of specialists by embedding their knowledge into the decision-making heuristics during the training process by augmenting the reward function with its expertise. This approach addresses critical societal challenges, including the global shortage of experts in high-demand fields like healthcare, law, etc. by scaling their expertise through AI-powered systems. By aligning AI development with domain-specific expertise, RLSF demonstrates the potential to deliver scalable, high-performing, and socially impactful solutions in fields where expert knowledge is scarce.

Project Goals

We want to show this framework beginning with a case study in the medical field comparing how much better an AI can do versus human baselines.

Duration

01/01/25 - 06/30/25

Project Lead and Team Members

Simon Lee (Ph.D. Student. Expected 2028)
Jeffrey N. Chiang (Ph.D.)
Sujay Jain (Undergraduate Student: Expected 2025)