NeutrinoReview 2.0: UI Redesign, System Stabilization, and Scientific Evaluation

This report details the development and research work undertaken on the NeutrinoReview application, an AI-powered tool to automate systematic reviews to conduct evidence based research during the 2025 CERN Summer Student Programme. The project began with a comprehensive overhaul of the system, including a complete re-architecture and redesign of all web components, significantly improving the application’s design and simplifying the user workflow. To modernize the development lifecycle, the application was containerized using Docker, a full CI/CD pipeline was established in GitLab, and the first version was successfully deployed on CERN OKD infrastructure, integrated with CERN Single Sign-On, and validated through user studies. Key algorithmic improvements were made, including debugging and enhancing the data deduplication module, creating a new data ingestion pipeline from ArXiv, and implementing a novel LLM Screening method called Cal-X. Additionally, research was conducted to validate the system's performance, with findings published in a research paper. Additional studies evaluated the reasoning capabilities of various Large Language Models laying the groundwork for future publication. This work has transformed NeutrinoReview into a robust, user-friendly, and scalable platform for scientific literature analysis.