Auditory verbal hallucinations are one of the most common symptoms in schizophrenia, contributing substantially to distress, functional impairment, and relapse risk. Existing therapeutic approaches, including pharmacological treatment and structured psychotherapies such as cognitive behavioral therapy for psychosis and AVATAR therapy, show promising clinical outcomes. However, scalability remains limited due to resource intensity and reliance on specialized personnel. Against this background, we describe the rationale, prototype architecture, and prospective validation roadmap for GOBLIN—a digital platform prototype combining personalized voice conversion, webcam-based eye-tracking, and large-language-model-assisted clinical documentation and decision support. The platform is intended to support clinician-supervised therapeutic workflows and between-visit monitoring following prospective validation. GOBLIN is intended to investigate whether an integrated digital workflow could address key limitations of current care models in schizophrenia and psychotic disorders, including fragmented follow-up, limited responsiveness to rapid symptom changes, and lack of scalable intervention tools. Our conceptual prototype is the first step toward leveraging the potential of digital, AI-driven platforms to complement existing treatment modalities and improve long-term management of psychosis-related symptoms. The system’s modules are intended for clinician-supervised research workflows and remain investigational. This manuscript does not report completed clinical validation, therapeutic efficacy, diagnostic performance, relapse-prediction performance, or validated safety outcomes. Such capabilities remain to be evaluated prospectively through staged technical, usability, safety, and clinical studies before clinical implementation can be considered.
This paper presents a criteria-governed, human-in-the-loop information system architecture integrating formative AI-assisted learning with preliminary assessment of spoken and written responses. It supports two modes: a retrieval-augmented assistant answering from course materials and an AI examiner scoring responses against a predefined rubric, criterion by criterion. The language model is confined to analysis: evidence extraction, aggregation, input-quality control, uncertainty estimation and escalation are performed by a controlled pipeline, and each result is stored as a structured decision object carrying criteria, evidence, provenance, uncertainty and teacher-verification status. Across 168 examination sessions and 1074 criterion-level decisions, every decision carried a verbatim evidence passage (100% coverage), the overall score matched the programmatic aggregation in all finalized sessions, 39.1% of responses were routed for secondary analysis and 5.4% of sessions were escalated. A survey of 52 students rated question clarity 4.81, criterion-level feedback 4.46, and allotted time 4.42, while concerns persisted about scoring reliability and recognition errors. In a preliminary blind calibration, the AI ranked 13 spoken answers like human raters (r = 0.75), but scored them 22 points lower. The contribution is the architecture and its feasibility in deployment; the study does not establish automated-scoring accuracy, and final decisions remain with the teacher.
Digital mental health interventions (DMHIs) frequently experience low user engagement and high attrition, highlighting the importance of incorporating user-centred design approaches during early-stage development. This study presents a structured mixed-methods workflow that integrates participatory co-design activities with a complementary quantitative prototype evaluation to support the formative development of digital mental health platforms. Using a qualitatively driven sequential design with an embedded quantitative strand, 45 university students in design, engineering, architecture and ergonomics programmes took part in three participatory co-design workshops in nine groups of five and, at the close of the final session, each group rated the low-fidelity AMGR-I prototype it had itself built against eleven design criteria. Ratings differed systematically across criteria (Friedman χ2(10) = 225.73, p < 0.001, Kendall’s W = 0.50), and the difference persisted when the clustering of ratings within the nine artefacts was modelled. Gamification was rated significantly lower than every other criterion after Holm’s correction and content quality highest, while data privacy and security was among the lowest-rated criteria despite being one of the most frequently and forcefully discussed concerns in the workshops. Because the instrument measured evaluation of the artefact rather than the importance attached to each criterion, that pattern is a convergence between the two strands rather than a paradox: the requirement these participants discussed most was the one their own artefacts served worst. That convergence is the study’s central integrated finding. The participants are characterised throughout as design-literate formative evaluators rather than as prospective users, and the resulting design propositions are presented as hypotheses for subsequent testing. The contribution of the paper is the workflow itself: a lightweight, low-cost and classroom-deployable procedure for combining co-design with structured prototype evaluation at the pre-clinical formative stage.
Cooling-tower noncompliance can weaken controls intended to limit Legionella-related public-health risks. We examined whether longitudinal inspection histories could predict at least one violation at a system’s next inspection and support capacity-constrained prioritization. We reconstructed 109,913 retained inspection-result records from the New York City Department of Health and Mental Hygiene, covering January 2017 through March 2026, into 35,825 inspections of 6162 systems. After excluding each system’s first inspection, we modeled 29,663 repeat inspections of 5668 systems using 19 history-only predictors. We compared logistic regression, XGBoost, and TabM under system-grouped random and chronological designs, assessed discrimination and probabilistic accuracy, examined model reliance using permutation importance, and retrospectively evaluated prioritization at 10%, 20%, and 30% capacity. The grouped-random test set contained 4217 inspections, of which 1295 were noncompliant; average precision ranged from 0.6695 to 0.6852 and AUROC from 0.7864 to 0.8003. At 10% capacity, the models captured 26.56–27.26% of noncompliant inspections and 42.33–42.64% of violations, with lifts of 2.65–2.72. Performance declined on 5421 chronologically held-out inspections (890 noncompliant): average precision ranged from 0.2645 to 0.2939 and AUROC from 0.6221 to 0.6490. At 10% capacity, and the models captured 18.76–22.02% of noncompliant inspections and 24.05–26.23% of violations, with lifts of 1.87–2.20. The models generally only modestly outperformed four simple history-based rankings. Permutation importance indicated strongest reliance on prior violation burden, inspection-history depth, and elapsed-time measures. Historical records supported retrospective prioritization, but performance weakened for later inspections. Prospective evaluation is needed before operational use; predictions concern recorded regulatory noncompliance, not Legionella contamination or direct public-health risk.
Currently, educators are looking for new techniques in the teaching–learning process. Previous research has presented various teaching strategies designed to motivate students to learn. Existing research has focused on the use of spatial world design in elementary and secondary education settings. Therefore, it is essential to analyze how these immersive environments, designed and programmed by teachers, provide insights into the motivational perceptions of university students in teaching programming courses. This study employed a single-group, fixed-order, within-subject design (N=41) to examine students’ reported motivation after traditional instruction and a subsequent Roblox-supported activity and also to see if there is any difference in motivation when using Roblox, considering gender. The present study used the RIMMS to measure motivation. To analyze motivation using the teacher’s material and Roblox, the Wilcoxon signed-rank test was used, and to analyze how gender affects it, the Mann–Whitney U test was used. The results revealed a statistically significant difference between both modalities (Z=−2.264, p=0.024, r=0.354, N=41). With regard to gender, the results indicated no statistically significant difference between males (N=31, Mean Rank =21.60) and females (N=10, Mean Rank =19.15), U=136.500, Z=−0.567, p=0.571 (exact p=0.580). The findings indicate that students reported higher motivation scores after the Roblox-supported activity than after the preceding traditional instruction. However, given the fixed-order design and the lack of formal measurement equivalence between the contextualized RIMMS versions, these differences should be interpreted as exploratory descriptive evidence rather than as evidence of an effect attributable specifically to Roblox.
