摘要
The rapid growth of digital information has made manual literature review and report compilation increasingly time-consuming, particularly for academic and technical research tasks. This paper presents the Multi-Agent AI Research Assistant, a collaborative system in which multiple specialised artificial-intelligence agents — responsible for planning, information retrieval, analysis, citation validation, and report drafting — work together under a coordinating orchestrator to automate the end-to-end research workflow. The system is implemented using the CrewAI multi-agent orchestration framework, an NVIDIA NIM-hosted Llama-3.1-70B-Instruct large language model, and a Streamlit-based web interface, and retrieves source material concurrently from arXiv, Semantic Scholar, CORE, and DOAJ. By decomposing a single research query into smaller, agent-specific sub-tasks, the system reduces manual effort, improves task-distribution efficiency, and produces structured, citation-verified technical reports more reliably than a single-agent or manual approach. Evaluation of the prototype on sample research topics shows improved workflow transparency, faster report turnaround, and consistent citation formatting compared with manual research compilation. The limitations of the current prototype — including computational cost, dependence on internet connectivity, and inter-agent communication latency — are also discussed, along with directions for future enhancement such as multilingual support, integration with additional academic databases, and adaptive agent learning.