Abstract

The increasing prevalence of psychological stress, emotional instability, academic burnout, and social anxiety among students has emerged as a major concern in modern educational ecosystems. Rapid academic competition, excessive digital engagement, social isolation, irregular lifestyle patterns, and continuously evolving performance expectations have significantly influenced student mental well-being across schools, colleges, and universities. Despite growing awareness regarding mental health, existing wellness monitoring approaches remain largely dependent on periodic counselling sessions, static psychometric questionnaires, and delayed manual intervention mechanisms, which often fail to provide continuous emotional assessment and real-time behavioural understanding. Consequently, early-stage emotional deterioration frequently remains undetected until it begins to affect academic performance, cognitive stability, interpersonal communication, and long-term psychological health.Recent advancements in artificial intelligence, affective computing, multimodal behavioural analytics, and large language models have created new possibilities for developing intelligent and context-aware wellness monitoring systems capable of understanding human emotional behaviour in a more adaptive and personalized manner. However, many existing AI-based wellness platforms still operate using isolated emotional indicators or static classification pipelines, limiting their ability to capture temporal behavioural patterns, contextual emotional variations, and dynamic conversational interaction. Human emotional wellness is inherently multidimensional and continuously influenced by behavioural history, emotional transitions, environmental context, and social interaction patterns. Therefore, modelling student wellness requires a multimodal and temporally aware inference framework capable of integrating heterogeneous emotional and behavioural signals into a unified contextual representation.To address these challenges, this paper presents MindCheck, an OpenLLaMA-powered multimodal student wellness monitoring framework designed for contextual emotional assessment, adaptive stress estimation, and intelligent conversational support. The proposed architecture integrates DeepFace-based facial emotion recognition, psychometric questionnaire analysis, longitudinal mood tracking, behavioural context modelling, and OpenLLaMA-driven conversational intelligence into a unified emotionally aware inference pipeline. Unlike traditional wellness systems that rely on isolated emotional observations, the proposed framework performs contextual fusion of multiple behavioural modalities to generate a more reliable and temporally consistent representation of student emotional state.The proposed system models wellness estimation as a context-aware multimodal inference problem in which emotional confidence scores, historical behavioural patterns, psychometric observations, and conversational interaction history are collectively utilized for adaptive stress estimation and personalized response generation. Mathematical formulations for emotional confidence fusion, temporal wellness propagation, contextual stress estimation, and conversational memory inference are introduced to improve analytical explainability, temporal consistency, and behavioural obustness. The OpenLLaMA conversational layer further enables emotionally adaptive and human-centered interaction by generating supportive responses conditioned on contextual emotional state and longitudinal behavioural memory rather than static rule-based templates. From a system design perspective, the proposed framework emphasizes modularity, explainability, lightweight deployment, scalability, and safe conversational AI integration suitable for educational environments. Experimental evaluation demonstrates that the proposed architecture can effectively support multimodal emotional inference, contextual behavioural understanding, adaptive conversational reasoning, and longitudinal wellness estimation while maintaining practical deployment feasibility and computational efficiency.The findings of this research highlight the growing importance of emotionally aware artificial intelligence systems in educational support environments and demonstrate how multimodal behavioural intelligence, affective computing, and conversational large language models can be integrated into a unified wellness intelligence framework capable of supporting scalable, adaptive, and human-centered student wellness monitoring. The proposed architecture establishes a strong foundation for future research involving personalized emotional intelligence systems, transformer-based behavioural modelling, privacy-preserving wellness inference, and next-generation AIassisted educational support platforms.