IRPA: A Retrieval-Augmented Framework for Intelligent Research Paper Analysis and Research Gap Detection

Abstract

Abstract—Manual literature review has become harder to sustain as the volume of published research keeps growing. A student or researcher preparing a review still has to open one PDF after another, work out what each paper is trying to do, note the methodology and findings, and keep track of where the gaps are — a process that is slow and easy to get wrong once the paper count rises. Search engines and citation databases help with finding papers, but they stop short of reading a paper on the user's behalf, and generic summarizers tend to flatten a paper into a shallow, abstract-level paragraph that drops the detail a reviewer actually needs. This paper describes the Intelligent Research Paper Analyzer (IRPA), a system built around Retrieval-Augmented Generation (RAG) that analyzes an uploaded academic PDF end to end. The pipeline splits a paper into sections, encodes each one with a sentence-transformer model, and stores the resulting embeddings in a vector database so relevant sections can be retrieved on demand. When a user asks a question or requests a summary, the retrieved sections are passed to a large language model together with a grounding prompt, which is intended to keep the generated answer tied to content that actually appears in the paper rather than left to the model's own recall. The same pipeline extracts objectives, methodology, key findings, limitations, and citation information in a structured form, and flags limitations that recur across a user's uploaded papers as candidate research gaps. The current work focuses on the design, implementation, and functional validation of the proposed prototype rather than a large-scale comparative benchmark evaluation. The system demonstrates the feasibility of integrating retrieval grounding, structured information extraction, citation analysis, and research-gap identification within a unified research-paper analysis workflow.