Advanced Semantic Analysis of Research Papers Using a Retrieval-Augmented Architecture

The exponentially increased number of academic papers published lately is raising significant challenges for researchers, requiring new ways for reviewing massive amounts of information, like automated systems capable of analysing the researched content.

Large Language Models (LLMs) represent an answer to this need, nevertheless, due to the hallucinations, while the research requires rigorous information, a large-scale adoption in this field remains a challenge.

Current scientific literature mentions Retrieval-Augmented Generation (RAG) as a solution (Upadhyay & Viviani, 2025) for LLMs hallucinations, nevertheless (Godinez, 2025) mentions also some difficulties these systems have in providing citations.

This paper addresses these gaps, proposing a scalable RAG architecture built on top of Azure services (Azure Blob Storage, AI Search, OpenAI) that leverages embeddings and multidimensional vector spaces for an advanced semantic analysis, extended with automation tools for accessing curated data (Meacham & Sharafzad, 2025) at the beginning of the flow and for consuming the synthetic answers during the final validations.

Using a progressive methodology, the study confirms the quality of the returned answers gradually, testing different LLM components and scenarios for obtaining the most accurate responses and providing citations for an increased reliability and trustiness of the overall architecture.

Furthermore, the proposed framework prioritizes cost efficiency by using accessible services that are sufficient for supporting the literature review automation while decreasing the articles synthesis process from hours to minutes (Godinez, 2025)