Apply by Feb. 15 for Penn PORTAL career development funding to advance Learning Health System science. Learn more.

Chart Hero

Chart Hero

AI-powered chart review

Project status

Pilot/study with results

Collaborators

Sri Adusumalli, MD, MSHP, MBMI 

Jeffrey Moon, MD, MPH 

Joy Iocca, MSN  

Susan Regli, PhD 

Srinivas Sridhara, PhD 

Jason Lubken 

Michael Becker 

Asaf Hanish, MPH 

Anna Schoenbaum, DNP, MS 

Gordon Tait  

Ameena Elahi, MS, MPA 

John Hesington  

Tessa Cook, MD, PhD 

Philipose Mulugeta, MD 

Bonmyong Lee, MD 

Opportunity

Clinicians spend significant time piecing together information from multiple places in the electronic health record (EHR) to inform decision-making about a patient’s care. This time-intensive work leads to cognitive burden, workflow inefficiencies, and delayed decisions.  

An EHR-friendly AI chat tool is a promising approach for sifting through the data to surface key information. Although EHR vendors and third-party solutions are moving into this space, robust options are not available today. 

Intervention 

Chart Hero is an Epic-integrated conversational AI assistant that gathers information from across the patient record and summarizes it in one view, transforming laborious chart review into a much quicker and easier task. Powered by a large language model, it searches available data sources across a patient's chart – images, scanned documents, notes, summaries, and more – to assemble relevant information. Chart Hero responses provide citations that link to data sources so users can reference them directly. 

Chat prompts can be saved for future use and shared with other EHR users for greater efficiency and collaboration. Chart Hero also has a built-in feedback tool so clinicians can easily flag issues areas for improvement as we continue to enhance its functionality. 

Impact

As of January 2026, Chart Hero is currently in the testing phase with over 50 users across inpatient and outpatient settings including Radiology, Nuclear Medicine, Infectious Disease, Cardiology, and Emergency Medicine. In its first three months, this tool has produced over 1,000 outputs. It has been relatively well received, with a net promoter score of 54 and largely positive feedback (68 percent).  

To date, Chart Hero is most frequently used to surface patient history (diagnosis and treatment history), prepare for an upcoming new patient visit (generating summaries or letters), or for general patient summarizations.  

Our team is currently evaluating how much time this tool saves clinicians and examining potentially useful applications like clinical trial screening. As we establish outcomes and continue to refine the technology, we intend to expand Chart Hero’s availability so that more people can benefit across the health system. 

Innovation Methods

Journey map

A journey map is a visualization of a user's process to accomplish a task. Journey mapping involves plotting user actions onto a timeline. Details on users' thoughts, emotions, and feedback are then added to the...

Journey map

In our experimentation phase with alpha users, we investigated the pain points associated with manual chart burden workflows in breast imaging radiology. For example, we learned that generating an outside second-read summary is a labor-intensive process of hunting across the patient’s chart. This led us to create a structured output for breast...

Journey map

A journey map is a visualization of a user's process to accomplish a task. Journey mapping involves plotting user actions onto a timeline.

Details on users' thoughts, emotions, and feedback are then added to the timeline to provide a holistic view of the experience or journey. Journey mapping will help you uncover what's working well in the current state and identify key pain points that need addressing.

You can build a journey map based on several users' observations, creating an archetype user journey, or you can use a template in real time as you conduct individual observations of users.

Download template

Journey map

In our experimentation phase with alpha users, we investigated the pain points associated with manual chart burden workflows in breast imaging radiology. For example, we learned that generating an outside second-read summary is a labor-intensive process of hunting across the patient’s chart. This led us to create a structured output for breast imaging radiologists; with one click the desired format and available data sources become available for radiologist review.

Vapor test

A vapor test offers a product or service that does not yet exist. Vapor tests will help you answer the question, "Does anyone want this?" and generate credible evidence for demand. Vapor tests require carefully...

Vapor test

For our initial testing, we intentionally left the capabilities of hero vague for early testers in order to see how the assistant would be leveraged.  Based on the questions asked, we could then refine the capability or add more safeguards.

Vapor test

A vapor test offers a product or service that does not yet exist. Vapor tests will help you answer the question, "Does anyone want this?" and generate credible evidence for demand.
 
Vapor tests require carefully designed soft landings to protect against poor user experience. An example of a vapor test would be showing a product or service on a website to see how many people express interest by clicking on it.
 
In this case, a soft landing would involve showing an "out of stock" or "not accepting new clients at this time" message when users click on the offering. 

Mini-pilot

High-fidelity learning can come from low-fidelity deployment. Mini-pilots will allow you to learn by doing, usually by deploying a fake back end. You might try a new intervention with ten patients over two days in...

Mini-pilot

We initially created several modes for Hero to explore different technical approaches to answering questions. Modes varied in terms of how users might specify the data to search for or how responses were generated. Based on early feedback, we were able to converge on a single mode that handled the majority of questions.

Mini-pilot

High-fidelity learning can come from low-fidelity deployment.

Mini-pilots will allow you to learn by doing, usually by deploying a fake back end. You might try a new intervention with ten patients over two days in one clinic, using manual processes for what might ultimately be automated.

Running a "pop-up" novel clinic or offering a different path to a handful of patients will enable you to learn what works and what doesn't more quickly. And, limiting the scope can help you gain buy-in from stakeholders to get your solution out into the world with users and test safely.