Intelligent Multi-Agent Research Paper Analysis and Automated Literature Survey Generation Using Large Language Models and Graph Retrieval-Augmented Generation

The increasing volume of scholarly publications makes it difficult to identify relevant studies, compare methods and findings, trace claims to primary evidence, and produce a coherent literature survey.This paper presents a design for an intelligent multi-agent framework that combines document processing, a knowledge graph, semantic vector retrieval, graph-aware retrieval, large language model (LLM) synthesis, and claim-level verification.The proposed system represents papers and research entities in Neo4j and stores page-aware textchunk embeddings in ChromaDB.Five logical agents-Discovery, Paper Analysis, Knowledge and Retrieval, Research Intelligence, and Synthesis and Verification-coordinate the workflow through a stateful orchestrator.Hybrid retrieval combines semantic similarity with graph relationships, while evidence records retain stable paper, chunk, section, and page identifiers.The generation stage is designed to produce structured paper summaries, comparative analysis, candidate contradictions, candidate research gaps, and a section-wise literature survey with traceable citations.The paper formalises the architecture, data model, retrieval strategy, verification controls, and evaluation protocol.The available source material does not provide a completed implementation audit or measured benchmark results; therefore, this manuscript reports a proposed system and reproducible evaluation plan rather than claiming experimentally demonstrated gains.The central research question is whether combining graph context, semantic retrieval, agentic decomposition, and explicit verification improves evidence coverage and citation support over simpler baselines without making latency and error propagation impractical.