Volume 19
Abstract: This research designed and developed an agentic multi-agent AI system for facial emotion recognition (FER), powered by Google's Gemini 2.5 Pro large language model. The study introduced an Agentic system comprising five agents: Input, Orchestrator, FER, Evaluator, and Output, which together manage the processing and analysis of facial images. The system uses Gemini 2.5 Pro's zero-shot learning to classify eight emotions without fine-tuning. The system was tested on 5,148 grayscale facial images, achieving a high level of accuracy. It excelled in recognizing clear emotions such as "Surprise" and "Happiness,” but struggled with subtler ones like "Contempt". Notably, the model appeared to be overconfident, as evidenced by high confidence scores even when the results were incorrect. In conclusion, this study shows the promise of advanced LLMs in agentic systems for applications in psychology, affective science, and medical fields. While these models improve automation and scalability, further work is needed to address calibration and bias for sensitive domains like mental health. Download this article: JISARA - V19 N3 Page 82.pdf Recommended Citation: Ani, C., Nguyen, T.L., (2026). An Emotional Analysis for Psychology, Affective Science, and Mental Health Using Agentic Multi-Agent AI Systems. Journal of Information Systems Applied Research and Analytics 19(3) pp 82-97. https://doi.org/10.62273/MCHM1528 | ||||||