Penn AI announces the 2026 Discovering the Future of AI award recipients
Penn AI is pleased to announce the first four awardees for the Discovering the Future of AI awards. Fifty-four competitive applications were submitted in response to the request for proposals, representing creative and bold ideas in research and education across Penn’s schools.
In addition to the four awards totaling $450,000, an additional 31 faculty applicants representing eleven schools received awards for high performance computing needs that will be supported by the Penn Advanced Research Computing Center for an estimated value of $852,000, bringing the total support to 1.3 million dollars. Access to high-performance computing will give Penn researchers the ability to run state-of-the-art AI models, analyze far larger and more complex datasets, and pursue bold, high-risk ideas, while removing the financial constraints that normally limit experimentation and discovery.
The Discovering the Future award is “designed to catalyze high-risk/high-reward research and education at the intersection of artificial intelligence and domain-specific scholarship for the benefit of society. This program moves beyond incremental advances to support research with the potential to transform a field of study and/or the educational experience through the innovative application of AI”.
We congratulate the following four awardees of grants from this program:
CASPER4D: Computer Assisted Surgical Performance Evaluation via Reconstruction
Daniel A. Hashimoto, MD, MSTR
Assistant Professor of Surgery, Perelman School of Medicine, University of Pennsylvania
Attending Surgeon, Hospital of the University of Pennsylvania
Active Surgeon, Penn Presbyterian Medical Center and Pennsylvania Hospital
Senior Fellow, Institute for Biomedical Informatics, Perelman School of Medicine
Affiliated Faculty, General Robotics, Automation, Sensing, and Perception (GRASP) Laboratory, School of Engineering and Applied Science
In collaboration with the School of Engineering and Applied Science
CASPER4D is a collaborative research project that applies artificial intelligence to improve surgical quality and patient outcomes through the analysis of standard surgical video. Led by Daniel A. Hashimoto and Kostas Daniilidis, PhD (School of Engineering and Applied Science), the project will develop AI models that reconstruct a four-dimensional (4D) representation of the surgical field—capturing anatomy, motion, tools, and evolving events over time.
By transforming routine laparoscopic and robotic surgical video into interpretable models of surgical activity, CASPER4D aims to assess technical skill and predict clinical risk in real time. The team will initially focus on a high-risk step in pancreatic cancer surgery—pancreaticojejunostomy during robotic pancreaticoduodenectomy—while developing methods that can generalize across procedures and care settings.
Ultimately, the project seeks to improve surgical training, standardize performance across hospitals, and reduce complications, costs, and disparities in care.
The Penn AI Pedagogy Initiative: Building Capacity for Meaningful and Responsible Adoption at Scale
Seiji Isotani, PhD
Associate Professor, Graduate School of Education, University of Pennsylvania
Faculty Director, Learning Analytics and Artificial Intelligence Program, Perelman School of Medicine, University of Pennsylvania
In collaboration with the School of Arts & Sciences
The Penn AI Pedagogy Initiative is a university-wide project designed to support responsible and effective use of artificial intelligence in teaching and learning by directly involving faculty and students in the design and testing of AI-enhanced educational practices.
Led by Seiji Isotani, an internationally recognized learning scientist and engineer, the initiative uses a co-design model in which interdisciplinary student teams and faculty partners collaboratively identify instructional challenges and develop AI-supported teaching strategies grounded in real classroom needs. Over its first year, the project will support AI-enhanced learning activities across 18 courses, spanning multiple disciplines and all four undergraduate schools.
A central outcome of the initiative will be the Penn AI Pedagogy Repository, a shared digital resource that documents and disseminates instructional models, tools, and materials developed through the project. Together, these efforts aim to establish a scalable, evidence-based framework for integrating AI into education in ways that deepen learning, preserve the human dimensions of teaching, and can be adapted by institutions beyond Penn.
Molecule 3D Structure‑Informed Science Agentic LLM
César de la Fuente, PhD, FRSB
Presidential Associate Professor, Perelman School of Medicine, University of Pennsylvania
Director, Machine Biology Group, Perelman School of Medicine, University of Pennsylvania
In collaboration with the School of Engineering and Applied Science
ApexMol is a research project that will develop an artificial intelligence system capable of reasoning about and designing biomolecules by integrating natural language with three-dimensional molecular structure. The project aims to move beyond traditional one-dimensional molecular representations by explicitly incorporating 3D geometry into large language models for scientific discovery.
The research team will create and openly release BioChemInstruct, a dataset of more than 12 million paired examples linking molecular structures with scientific text from sources such as the Protein Data Bank, PubChem, UniProt, and AlphaFold predictions. Using this dataset, the project will train a unified agentic large language model that treats text and 3D molecular information as a single sequence, enabling the system to answer scientific questions, predict molecular properties, and design new compounds.
The model will be evaluated on core scientific tasks including molecule generation, binding-pose estimation, and drug repurposing using established benchmarks. By open-sourcing the data, models, and tools developed through the project, ApexMol seeks to accelerate drug discovery, support responses to emerging health threats, and broaden access to advanced AI capabilities for molecular science across academia, industry, and education.
Led by César de la Fuente, whose work spans artificial intelligence, biology, and medicine, the project reflects Penn AI’s goal of supporting high-risk, high-reward research with the potential to transform scientific practice through innovative applications of AI. is a collaborative research project that applies artificial intelligence to improve surgical quality and patient outcomes through the analysis of standard surgical video.
EchoMFM: A Multimodal Foundation Model for Automated Clinical Interpretation of Echocardiograms
Julio Alonso Chirinos-Medina, MD, PhD
Professor of Medicine (Cardiovascular Medicine), Perelman School of Medicine, University of Pennsylvania
Attending Physician, Hospital of the University of Pennsylvania
Co-Director, Training Program in Cardiovascular Biology and Medicine (Clinical/Translational Science T32), Perelman School of Medicine
Adjunct Faculty, Center for Magnetic Resonance and Optimal Imaging, University of Pennsylvania
Adjunct Faculty, University of Ghent, Belgium
EchoMFM is a research project that will develop a unified, multimodal artificial intelligence system to support the clinical interpretation of echocardiograms by integrating imaging data with complementary clinical information. The project aims to reduce variability in interpretation, improve diagnostic accuracy, and increase efficiency in cardiovascular care.
The system will integrate echocardiographic imaging with electronic health records, cardiology reports, electrocardiograms, and cardiac MRI data to train a foundation model that learns shared representations across modalities. Using these representations, EchoMFM will be adapted to perform key echocardiography tasks, including disease classification, anatomical segmentation, and quantitative measurement, with particular attention to complex and underrepresented cardiovascular conditions.
A core feature of the system will be the generation of both structured outputs and draft narrative reports, which clinicians can review and finalize rather than produce manually. To promote transparency and clinical trust, the model will also highlight the specific images and views that support each reported finding. The project is designed with real-world adoption in mind, with planned integration into clinical systems such as Epic.
Led by Julio Alonso Chirinos-Medina, a cardiovascular physician-scientist whose research focuses on arterial stiffness, pulsatile hemodynamics, and heart failure with preserved ejection fraction, EchoMFM reflects Penn AI’s goal of advancing high-risk, high-reward research that translates artificial intelligence into meaningful clinical impact.