<![CDATA[Mental health disorders among adolescents in Indonesia remain largely underdetected due to limited access to services, persistent stigma, and the lack of personalized feedback in conventional screening tools. This study developed and evaluated a web-based mental health screening system that integrates the DASS-21 questionnaire with a large language model (GPT-4) to generate personalized intervention recommendations. The system was built using the Waterfall methodology and designed to calculate DASS-21 severity scores for depression, anxiety, and stress, then pass both quantitative scores and optional user free-text input to the LLM via API. Black-box testing was conducted to validate functional requirements, and the System Usability Scale (SUS) was administered to 30 adolescent users to assess usability. Results showed that all functional test cases passed after resolving an initial LLM response parser issue. The average SUS score was 91.59 (Grade A, Acceptable range), with no participant rating the system below 70, indicating consistently high usability across users. The hybrid approach proved advantageous: the DASS-21 provided clinical grounding that reduced LLM hallucination risk, while the LLM added contextual personalization that static questionnaires lack. However, the high usability score does not automatically translate to clinical effectiveness. Future work should include clinical validation studies comparing LLM-generated recommendations against psychologist assessments.]]>
Copyrights © 2026
