AI RESEARCH PAPER ASSISTANT

Abstract

The growing volume of scientific publications makes it difficult for students and researchers to locate and understand relevant information efficiently. This paper presents the design of an AI Research Paper Assistant, a web-based application intended to support the analysis of research documents through PDF text extraction, document summarization, semantic retrieval, and natural-language question answering. The proposed workflow extracts text from an uploaded document, divides the content into manageable passages, converts passages into vector representations, and retrieves relevant sections in response to user questions. A Retrieval-Augmented Generation (RAG) approach can then use the retrieved passages as context for generating document-grounded responses. The planned implementation uses a React-based interface, a Python FastAPI backend, PyMuPDF for PDF text extraction, a sentence-embedding model, ChromaDB for vector storage, and a locally served language model through Ollama. The application is intended to reduce repetitive manual searching and make research content easier to explore. Its reliability must be assessed through experiments measuring retrieval relevance, answer correctness, citation support, response time, and usability. This paper describes the proposed design and evaluation approach; quantitative performance claims are reserved for results obtained from actual testing.