Serapis: a generative AI model to produce social media intelligence for depression diagnosis

Depression affects millions worldwide, generating multidimensional burdens. Scalable and timely diagnosis remains challenging. This study presents Serapis, a computational model that assists mental health professionals by converting social media content into structured intelligence for depression diagnosis. The model integrates large language models (LLMs) with the Structured Personal Mental Health Ontology (SPMH), developed using principles from forensic computing and Open-Source Intelligence (OSINT). Serapis collects data from consented social media posts, structured questionnaires, and Patient Health Questionnaire-9 (PHQ-9) responses. The system classifies each post, applies data augmentation to infer user traits, and constructs a timeline highlighting diagnostically relevant information. A Minimum Viable Product (MVP) enabled iterative development and evaluation with five psychologists. Focus group sessions and Technology Acceptance Model (TAM) supported the assessment of usefulness and usability. Participants qualitatively reported increased diagnostic confidence and perceived potential improvements in workflow efficiency and interpretability, underscoring Serapis' practical value for clinical reasoning. Serapis enforces consent-based data handling and protects patients' autonomy. The results demonstrate how generative artificial intelligence and social media intelligence can contribute to secure, modular, potentially scalable, and clinically meaningful support for depression diagnosis. The model delivers structured, reproducible insights through transparent, ontology-driven interpretation of digital content.