An Empirical Evaluation of a Multi-Agent Framework for Retrieval-Augmented Academic Research
Vaibhav Hingnekar · 2025
The exponential growth of academic literature presents a significant challenge for researchers in information discovery. This paper presents and evaluates Second Mind AI, a novel framework that functions as an intelligent research assistant through a modular multi-agent architecture. The system synergizes the generative capabilities of Large Language Models (LLMs) with factual data retrieved in real-time from the Semantic Scholar academic graph API. We detail the system's architecture and its core methodology, which features a unique two-phase iterative refinement mechanism. An empirical evaluation was conducted to measure the system's performance against two relevant baselines: manual search on the Semantic Scholar website and a general-purpose LLM. Our results demonstrate that Second Mind AI significantly reduces the time required to find relevant literature and completely eliminates the issue of source hallucination present in the baseline LLM. These findings validate our framework as an effective tool for improving the efficiency and reliability of academic research workflows.