On 2–3 October 2026, the University of Information Technology (UIT, VNU-HCM) hosted HealthInfo IV, the 4th International Health Informatics Conference. I was invited to speak at the Health AI Innovation Forum about psychotherapy, and about the work we are doing at psylab.
The one treatment that leaves no record
Health AI has already reached imaging, medical records, logistics and medicine itself. Psychotherapy is different. It is delivered entirely in conversation, and in most places that conversation leaves no record. The clinical note is written from memory. Supervision works from the therapist’s own account of what happened. And a service usually cannot say whether patients actually received the treatment they were meant to receive.
Therapy is a skill, and therapists differ
The research on this is clear. In routine services, the least and most effective therapists can differ substantially in how many of their patients recover (Saxon & Barkham, 2012). Years of practice alone do not seem to improve outcomes (Goldberg et al., 2016), and supervision, as it is usually practised, explains very little of the difference (Rousmaniere et al., 2016; Whipple et al., 2020).
What does help is working with the session itself: rehearsing concrete interventions rather than only discussing them, and practising difficult moments with expert feedback. That is hard to do when supervision depends on recall, and when, as one classic study found, most trainees hold something back from their supervisors (Ladany et al., 1996).
Why the instruments stay in research
Validated tools for rating how well a therapy was delivered already exist. In mentalization-based treatment (MBT), for example, there is the MBT Adherence and Competence Scale. The problem is time: a trained rater needs about an hour per session, so this kind of feedback rarely leaves the research setting.
Why now, in Vietnam
Vietnam is building its psychotherapy workforce and its standards right now. Clinical psychologist became a licensed title in 2024, the treatment gap is very large, and national guidelines for psychological interventions are being developed. Once a treatment is clearly defined, it becomes possible to ask whether it was delivered as defined.
What psylab is building
At psylab, we are exploring what changes when the session itself becomes the record. From a recorded session, the system produces a transcript and an expert-style reading, scored on items adapted from the MBT-ACS. It shows where the session followed the method and where it drifted, and every score links back to the exact moment in the transcript.
The therapist sees the reading first and decides what to share and with whom. The aim is not to replace supervision, but to let the supervisor’s time go to the moments that most need a person.
What we don’t yet know
I tried to be honest about the limits. Our early internal checks are encouraging but small. We do not yet know how the system compares with agreement between human experts, how it performs in Vietnamese, how it holds up across many sessions and therapists, or whether the feedback ultimately improves patient outcomes. Those are the questions that matter most.
An invitation
A health system that is building its psychotherapy workforce now does not need to inherit the unmeasured model. Getting there would take recording and consent practices that clinicians and patients accept, a rating scale validated in Vietnamese, supervisors trained to work from readings, sound data governance, and research that tests the link to patient outcomes. None of this is something one company can do alone.
If you work in a clinic, a training programme, research or health policy and are interested in this, I would be glad to hear from you at hieu@psylab.ai.
Thanks you to the organisers of HealthInfo IV for the invitation.