Transcript-derived patient profiles conditioning virtual patient behavior across seven dimensions¶
Design · All designs
Evidence · no claims cited
Description¶
Four anonymized patient profiles (two male, two female) were synthesized with GPT-5.2 from 4–7 real ACT sessions each, which "averaged 58–92 therapist-patient turn pairs and lasted 38–60 minutes". A structured prompt extracted stable patterns across seven dimensions (patient overview, core difficulties, values and motivations, emotional and cognitive themes, behavioral patterns, progress trajectory, simulation hooks) and constrained the model to rely only on transcript content and produce anonymized descriptions.
Design Implications¶
Context¶
Requirements¶
- Real therapy transcripts as source material
- Joint analysis of multiple sessions to capture stable patterns
Constraints¶
- Profiles are anonymized approximations derived only from transcript content
Target Learners¶
- Psychotherapists in training
Learning Goals¶
- Clinically plausible patient interaction practice
Claims¶
Related Designs¶
- Four-stage scenario generation pipeline with demographic sampling, quality verification, and adversarial extension packs
- Adaptive Virtual Patient framework: separating behavioral state transitions from language generation
Examples¶
-
Key Sources¶
- Pascal Riachi, Sofie Kamber, Stella Brogna, Andrew Gloster, Rafael Wampfler. (2026). Toward Accessible Psychotherapy Training Using AI-Driven Interactive Patient Avatars. https://arxiv.org/abs/2606.17786
Page metadata
| Field | Value |
|---|---|
type |
design |
id |
transcript-derived-virtual-patient-profiles |
title |
Transcript-derived patient profiles conditioning virtual patient behavior across seven dimensions |
description |
Four anonymized patient profiles (two male, two female) were synthesized with GPT-5.2 from 4–7 real ACT sessions each, which "averaged 58–92 therapist-patient turn pairs and lasted 38–60 minutes". |
status |
review |
generated |
by process:settle-candidates on 2026-10-10 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — pascal-riachi-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.