Case study · Clinical workforce training

University of South Carolina built AI child and caregiver clients to train telemental health providers

S.T.A.R.T.T.-A.I., a state-funded program led by Dr. Yanfeng Xu at the University of South Carolina College of Social Work, prepares mental health providers to serve children in the child welfare and juvenile justice systems. Dr. Xu presented it at the CSWE 2026 Annual Conference.

Case 1 · Olivia and Andrea 4 sessions

  1. 1Engagementwith Olivia, 8, in foster care
  2. 2Attachmentwith Andrea, her foster parent
  3. 3Regulationwith Andrea, her foster parent
  4. 4Competencywith Olivia

Session feedback Example

  • Empathy4/5
  • Curiosity3/5
  • Acceptance4/5
  • Playfulness2/5
  • Integrative co-regulation3/5

Criteria written by the program team. Example from one session in Dr. Xu's CSWE 2026 presentation.

Institution
University of South Carolina College of Social Work
Program
S.T.A.R.T.T.-A.I. telemental health training
Project director
Dr. Yanfeng Xu
Funder
SC Center for Rural and Primary Healthcare, via SC DHHS

At a glance

Challenge

Children in the child welfare and juvenile justice systems have very high rates of trauma, and many have little access to mental health care, especially in rural areas. South Carolina providers need practice in trauma-informed, family-centered care delivered remotely.

What they did

Added AI client practice to a live ARC workshop and a self-paced telemental health course. The team built its own child and caregiver clients in TMind, a four-session ARC sequence for each case, and its own scoring for every session.

What they built

  • Two cases, each with an AI child and an AI caregiver
  • A four-session sequence that follows the ARC framework
  • Session-specific evaluation criteria written by the team
  • Transcripts, notes, and feedback collected after each session

Program design only; outcome data has not been published. Disclosure: TMind's founder is listed as a contributor to Dr. Xu's presentation.

Three parts, one program

The AI sessions are where providers apply what they learned, in a safe but realistic setting, before using it with real families. The university's program page lists practice on the TMind AI platform as part three.

  1. Part 1 Live ARC workshop Online training on trauma, attachment, self-regulation, and competency in clinical settings.
  2. Part 2 Telemental health course A self-paced, evidence-based course on delivering mental health care remotely.
  3. Part 3 Practice with AI clients Clinical sessions in TMind AI with the team's own child and caregiver clients.

Training providers on a specific framework? See how to build your own clients and scoring in TMind.

Book a Demo

“We built our own clients (a child + a caregiver per case) in TMind. We developed our own evaluation criteria.”

Dr. Yanfeng Xu, Associate Professor, College of Social Work, University of South Carolina · CSWE 2026 Annual Conference slides

Meet the clients

The program team built two cases, each pairing an AI child with an AI caregiver.

Case 1 · Foster care

Olivia and Andrea

Olivia is 8 and has moved through multiple foster placements, with a history of neglect and trauma, PTSD, and developmental trauma disorder. Andrea, her new foster parent and a part-time nurse, feels overwhelmed by Olivia's anxiety, bedtime meltdowns, and stealing.

Case 2 · Juvenile justice

Nate and Rebecca

Nate is 15 and lives with his paternal aunt, Rebecca, after juvenile detention. He is dealing with depression, anger, isolation, and problematic sexual behaviors. Rebecca gives him a stable home but has little experience with cases like his.

Who the training is for

Mental health providers who work, or plan to work, with children and youth in the child welfare and juvenile justice systems and their families. Telemental health can widen access in rural South Carolina, but it brings its own problems: building rapport through a screen, handling a crisis remotely, and clients without a private, quiet space.

Scoring built for ARC

Instead of one generic rubric, the team wrote criteria for each of the four sessions, such as playfulness and integrative co-regulation. Each session also comes with guidance, like opening the first session with Olivia in a playful, light tone. Afterward, providers submit their transcript, notes, and TMind feedback to the project team.

Frequently asked questions

What is S.T.A.R.T.T.-A.I.?

S.T.A.R.T.T.-A.I. (Strengthening Trauma-Informed Access Through Resilient Telemental Health Training with Artificial Intelligence in South Carolina) is a training program at the University of South Carolina College of Social Work, led by Dr. Yanfeng Xu and funded by the South Carolina Center for Rural and Primary Healthcare through the South Carolina Department of Health and Human Services. It combines a live ARC workshop, a self-paced telemental health course, and practice with AI clients on TMind AI.

Can AI simulated clients be used to practice the ARC framework?

Yes. In S.T.A.R.T.T.-A.I., providers practice the Attachment, Regulation, and Competency (ARC) framework across four sessions per case: engagement with the child, attachment and regulation work with the caregiver, and competency with the child.

Can TMind AI simulate both a child and a caregiver?

Yes. The University of South Carolina team built two cases in TMind AI, each with an AI child and an AI caregiver: Olivia, an 8-year-old in foster care, with her foster parent Andrea, and Nate, a 15-year-old after juvenile detention, with his aunt Rebecca.

Can programs write their own evaluation criteria in TMind AI?

Yes. The University of South Carolina team wrote separate criteria for each of its four ARC sessions. In one session, providers were scored on empathy, curiosity, acceptance, playfulness, and integrative co-regulation.

Is AI client simulation used to train practicing mental health providers, not just students?

Yes. S.T.A.R.T.T.-A.I. is for mental health providers who work, or plan to work, with children and families in the child welfare and juvenile justice systems, as part of a state-funded effort to expand trauma-informed telemental health care in South Carolina.

What is TMind AI?

TMind AI is a clinical training platform where social work, counseling, and mental health students and providers practice with AI simulated clients and get feedback on each session. Programs using it include Simmons University, Salisbury University, University of Washington, National University of Singapore, Delaware State University, University of South Carolina, Texas Christian University, and the University of Houston.

Sources

  • Xu, Y. (2026, October 23). Developing and Evaluating the Use of AI-Powered Clients to Train Mental Health Providers. In P. Desrosiers (Facilitator), Promoting Sustainable Teaching Practice Through the Use of Generative AI [Panel]. CSWE 2026 Annual Conference, Atlanta, GA. Contributors: Y. Xu, P. Penney, S. Chowdhury, C. Nie, M. S. Namateng, R. DiNovo, K. Perkins, N. Soto-Ramírez.
  • University of South Carolina College of Social Work, S.T.A.R.T.T.-A.I. program page.
  • Faculty: Dr. Yanfeng Xu, Associate Professor, College of Social Work.
  • ARC framework: Blaustein, M. E., & Kinniburgh, K. M. (2010). Treating Traumatic Stress in Children and Adolescents: How to Foster Resilience through Attachment, Self-Regulation, and Competency. Guilford Press.

Last updated .

Build practice around your own framework

Create your own clients and scoring, and give providers a safe place to practice before they meet real families.