Children’s Hospital of Philadelphia is deepening its investment in clinician-facing AI with the development of CHIPPER, a virtual assistant embedded within its Epic EHR platform.
Designed to streamline time-consuming, cognitively demanding tasks for providers, CHIPPER is currently in technical development and aims to become a clinical co-pilot capable of synthesizing data from internal systems and external, evidence-based sources.
“The opportunity with AI is that we now have the ability to let the AI do some of that fetching and information synthesis and summarization for you,” Bimal Desai, MD, vice president and chief health informatics officer at CHOP, told Becker’s. “That’s what language models are especially good at.”
Unlike consumer-facing AI models, CHIPPER is being architected to avoid common pitfalls such as hallucinations or unsafe recommendations. CHOP’s team designed the system to only retrieve and summarize information from trusted clinical sources.
“We explicitly architected this to take advantage of that notion — that we have trusted datasets and resources that we want people to refer to,” Dr. Desai said. “We have the CHOP Clinical Pathways website, our internal policies and procedures, and third-party materials like UpToDate and VisualDx.”
CHIPPER uses a framework called “Model Context Protocol,” which allows the system to route specific types of queries to corresponding trusted APIs. For example, a request about drug interactions is sent to a medication reference database, while a question about evidence-based alternatives may query PubMed.
In demonstration scenarios, CHIPPER is already capable of retrieving a patient’s current medications from Epic and displaying them in a clean, formatted table — before escalating the interaction to suggest safer alternatives based on pediatric evidence.
While the tool has yet to be deployed in a clinical setting, Dr. Desai said the development team is focused on proving its technical feasibility and reliability across different sources.
“We’re still a ways away from rolling this out to the masses,” he said. “At this stage, I want to prove the technical feasibility, the architecture, the reliability of the resources, etc. It’s meant to be provocative — I want people to say, what are the downstream reference resources we should connect to CHIPPER that would be most helpful?”
CHOP plans to follow a phased rollout strategy, modeled after other AI implementations at the health system. Earlier this year, CHOP scaled an Epic-supplied AI feature that summarizes outpatient visit notes — first testing it with a handful of users in a non-production environment, then gradually expanding access after rigorous evaluation and feedback collection.
“That tool generated more than 20,000 summaries and analyzed over 100,000 clinical notes before we deployed it enterprise-wide,” Dr. Desai said. “We’ve learned how to do this safely and effectively.”
The project has also revealed key lessons about the unique challenges of deploying AI in pediatric environments. From differences in clinical models to additional complexities in patient-clinician interactions, CHOP’s team is building with these nuances in mind.
“In a pediatric setting, I’ve got the patient, the provider, the sibling, the parent — multiple participants in the room,” he said. “Even something like distinguishing voices correctly in ambient documentation becomes much harder. These tools need to be tested and validated locally.”
CHIPPER was initially built by Stephon Proctor, PhD, a board-certified clinical psychologist and associate professor of clinical psychology in psychiatry at the University of Pennsylvania, who Dr. Desai described as “a self-professed full stack informaticist” despite not having a traditional coding background. Using generative AI tools to write code, Dr. Proctor rapidly prototyped the first version of CHIPPER.
“We are in this very cool era where people with the right idea and right expertise can rapidly prototype tools,” Dr. Desai said. “Then the question becomes, how do you make this production-grade?”
To centralize innovation efforts, CHOP recently created a new leadership role — associate chief health informatics officer for EHR platform and innovation — and appointed Dr. Proctor to the post. The position is designed to help CHOP rapidly assess, test and roll out emerging AI tools within clinical workflows.
“In order for a health system to get good at this kind of thing, you really need a core focus on R&D,” Dr. Desai said. “We need the muscle, the infrastructure, the evaluation expertise — everything it takes to do this well.”
Though CHIPPER is still in its early stages, Dr. Desai said the work is laying the foundation for a more intelligent, clinically integrated future.
