Existing literature on virtual agents shows strong progress in conversational AI but still lacks emotionally responsive, multilingual, and context-adaptive systems capable of sustaining long-term engagement across varied domains. Most current models remain text-centric and do not support real-time emotional reciprocation or cross-commodity use. To address this gap, the study presents Petzy, a 3D AI-powered virtual pet designed for friendly, low-pressure interaction. Its dual-zone behaviour design—Fun Zone and Intellectual Zone—supports emotional computing and adaptive engagement. Petzy identifies user emotions through speech and language analysis and responds using a multilingual intelligent chatbot paired with dynamic behavioral traits such as affection, vitality, thirst, and happiness. The system is built using React JS, Three JS, Blender for 3D modelling, Firebase for secure cloud storage, and FastAPI for backend routing, with Groq API and Ollama enabling high-performance LLM inference and multilingual speech detection. Experimental trials assessed latency, emotional accuracy, adaptive learning, and engagement. Results show Petzy delivers low response times, stable emotional feedback, and high user participation while functioning as both an emotionally aware companion and a domain-specific simulator for training and learning contexts.
M. et al. (Fri,) studied this question.