International Journal of Scientific Research and Engineering Development

The rapid advancement of artificial intelligence has led to the development of highly capable conversational agents. However, most existing systems remain reactive, single-domain, and incapable of autonomously executing complex real-world tasks. This limitation restricts their applicability in dynamic, multidomain environments. To address this challenge, this study proposes NOVA, a multi-agent agentic AI assistant designed for autonomous task execution across diverse domains. NOVA integrates intent classification, task orchestration, and domain-specific agents into a single framework. The system employs a hybrid architecture in which user input is processed through an intent detection module, followed by an orchestrator that dynamically routes tasks to specialized agents, such as website generation, shopping automation, and research assistance. The implementation leverages machine learning models, including DistilBERT, for intent classification and utilizes external tools and APIs for execution. Experimental evaluation demonstrated improved task completion rates, with intent classification accuracy reaching 90–92% and domain-specific task success rates exceeding 80%. A comparative analysis with baseline chatbot systems highlighted NOVA’s superior performance in multistep task execution and contextual understanding. The results validate the effectiveness of agentic architectures in enabling scalable and autonomous AI systems.

Kunal Singh, Samarth Negi, Shubhanshu Singh Fartyal, Jyoti Gaur,"NOVA: A Multi-Agent Agentic AI Assistant for Autonomous Task Execution Across Domains" International Journal of Scientific Research and Engineering Development, V9(2): Page(2238-2243) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.