PeerCoPilot Uses AI To Support Community Mental Health Care
Marylee WilliamsTuesday, February 24, 2026Print this page.
The Breakdown
- PeerCoPilot is a chatbot built to support peer-led mental health services.
- The tool uses vetted service databases and public benefits documentation to reduce AI inaccuracies.
- CMU researchers worked with frontline peer support specialists to ensure their work reflects real-world workflows and needs.
***
Carnegie Mellon University researchers have developed an AI-powered chatbot designed both with and specifically for people working in behavioral health.
The chatbot, known as PeerCoPilot, uses only vetted information to help mental health support workers assist people with their wellness needs, create actionable plans and check benefits eligibility.
"The stakes in this work are important," said Naveen Raman, a Ph.D. student in the School of Computer Science's Machine Learning Department (MLD) who worked on PeerCoPilot. "In addition to the technical challenges, there's also a human element, ensuring that the people who work with this tool know how to use it. As a result, we need to make sure we build in extra lines of defense against inaccurate information by combining technical guardrails with human training."
PeerCoPilot provides narrowly focused responses to mental health support questions. The tool can provide wellness plans and resource recommendations, check benefits eligibility, or ask follow-up questions. If someone needs housing but lacks an ID needed to apply, for example, PeerCoPilot will check the person's eligibility, find nearby resources to help them find temporary shelter, and create a plan to address identification and other needs.
To reduce errors, PeerCoPilot uses retrieval-augmented generation. Current large language models pull the information for their answers from various locations, which can result in incorrect answers if the vat the LLMs pull from isn't curated. PeerCoPilot only pulls information from vetted sources, such as verified databases of local community services, publicly available benefits and eligibility documentation, and organization-approved resource lists.
Researchers worked with staff at Collaborative Support Programs of New Jersey (CSPNJ), a peer-led, not-for-profit organization, to design and test PeerCoPilot. Peer-run organizations play a vital role in helping people who experience behavioral health and other disorders, such as substance use. These organizations offer a safe environment for people most in need to access care, from obtaining a license to determining how to get a good night's sleep, and they're run by individuals who have experienced similar issues.
"What really makes this partnership successful is that we are taking a participatory approach to AI design that centers on the impact from and on the communities these peer-run organizations serve," said Hong Shen, an assistant professor in the Human-Computer Interaction Institute (HCII). "We are using a lot of methods from human-computer interaction to empower the peer support providers in this process, no matter what technical background they may have. Our goal is to maximize AI's benefits while minimizing its risks, and participation from impacted communities is at the heart of that."
For the last two years, CMU researchers worked with CSPNJ peer staff to gather information for designing and testing PeerCoPilot.
"This project has been a learning experience for everyone on our end," said Peggy Swarbrick, CSPNJ's Wellness Institute Coordinator. "It has allowed us to learn about this rapidly evolving technology. We hope this will help people practically."
Researchers plan to continue developing PeerCoPilot and are in talks with other community support organizations to see how they might customize the tool for widespread use.
Along with Shen, Raman and Swarbrick, the PeerCoPilot research team included Gao Mo, a former SCS research associate; Megan Chai, an HCII alumna; Jini Kim, an HCII doctoral student; Cindy Peng, an HCII research associate; Olivia Cheng, a master's student in MLD; Ashray Pamula, an SCS undergraduate student; Ningjing Tang, a HCII doctoral student; and Fei Fang, an associate professor in the Software and Societal Systems Department. Researchers from the University of Pittsburgh's School of Social Work were also involved, including Shannon Pagdon and Nev Jones.
The National Science Foundation's Civic Innovation Challenge grant supported this work, which was accepted at April's Association for Computing Machinery Conference on Human Factors in Computing Systems (CHI 2026).
Aaron Aupperlee | 412-268-9068 | aaupperlee@cmu.edu