Web Application for Scientific Paper Retrieval and Summarization Using Large Language Models

This research addresses the challenge of effectively discovering and synthesizing scientific literature, a growing concern given the exponential increase in publication volume. To improve search relevance and streamline academic workflows, we developed a web application that leverages semantic search using the all-mpnet-base-v2 embedding model combined with pgvector indexing (Hierarchical Navigable Small World). The system supports automatic paper summarization, citation formatting across multiple styles, similarity graph visualization, and retrieval-augmented generation (RAG) for related work writing. The backend architecture is built on FastAPI and PostgreSQL, with a React-based frontend. Performance was evaluated using standard information retrieval metrics including Recall@3, MRR@3, Precision@3, and MAP@3. The embedding-based approach consistently outperformed a traditional TF-IDF baseline across all measures, confirming its ability to retrieve more contextually relevant results. These findings demonstrate the practical value of integrating large language models and vector-based retrieval in academic tools. The system contributes to ongoing efforts to enhance literature review processes and could serve as a foundation for more advanced, AI-assisted research environments.