This abstract was part of the 2026 UCSF AI and Education Symposium
Background and Educational Need
High-quality feedback is central to competency-based medical education, promoting self-reflection, identifies performance gaps, and provides the opportunity to drive behavior change, yet achieving it consistently in clinical environments remains difficult. Although systematic reviews suggest that longitudinal feedback improves trai
nee performance, the delivery of effective feedback between trainees and faculty remains inconsistent due to limited time, workflow interruptions, and documentation fatigue. Evaluations of workplace-based assessments across multiple residency programs have found that narrative comments are often too brief or general to guide meaningful improvement. These limitations highlight an opportunity to use voice-enabled AI to reduce the burden of generating evaluations and to streamline the production of high-quality feedback, thereby strengthening professional development.
Solution and Prototype
To overcome the aforementioned barriers, we deployed a commercial AI platform, FanVoice (developed by FanKave Inc., a company specializing in AI-driven solutions), at UCSF to collect, collate, and generate narrative summaries of audio and text-based feedback during the cardiology fellowship. FanVoice is a voice-first generative AI tool that structures and synthesizes brief verbal responses to support more timely and detailed feedback. We have tailored Fanvoice to process audio or video-recorded feedback and produce structured, actionable summaries aligned with predefined evaluative domains. The platform will also consolidate input from multiple evaluators, enabling the creation of aggregated insights that highlight recurrent themes while preserving anonymity. Trainees and faculty will have access to their individualized feedback via a customizable interface within FanVoice. Program leadership may additionally query the aggregated dataset to explore specific competencies or patterns, supporting more efficient and data-informed assessment oversight. Importantly, the platform’s contextual-analysis engine can identify feedback expressing significant concern related to trainees, faculty, or rotations, that would allow for rapid and real-time corrective attention.
Feasibility Testing and Evaluation Plan
The pilot implementation plan has been deemed by the UCSF IRB as not human subjects research and qualifies as a quality improvement initiative. A small group of UCSF faculty and trainees performed beta testing to assess functionality, usability, and accuracy of the platform. The early testing phase will begin in January 2026, with a plan to implement the platform on three rotations within the cardiology fellowship program at the San Francisco VA Medical Center. After initial testing, we will plan to expand platform usage to additional rotations at all three UCSF clinical sites and ultimately expand to additional training programs. We plan to collect data in a systematic, prospective fashion to evaluate the feasibility and efficacy of the intervention. We will track metrics including evaluation completion rates, time-to-completion, and structured user satisfaction surveys, which will be compared with equivalent data from the MedHub platform. Program leadership will also have access to deidentified feedback to further test the capacity of FanVoice to provide aggregated insights.
This voice-first generative AI platform provides a scalable path toward improving the quality, timeliness, and completion rate of trainee–faculty feedback thus fostering a clinical learning environment that is genuinely fun, fast, and formative.
Contact
Davis Kimaiyo, [email protected]
Jessica Holtzman, [email protected]