Advancing NeutrinoReview: Creating an End-to-End Platform for Systematic Reviews with Large Language Models and Automated Full-Text Retrieval

About

While systematic reviews (SRs) are usually the most comprehensive method for scientific evidence synthesis, they require researchers to invest considerable time and effort. The NeutrinoReview platform is a proof-of-concept application to automate the labor-intensive steps of SRs by integrating LLM screening techniques like the 5-Tier and CAL-X algorithm, focused on reducing human workload while preserving high sensitivity.

This project extends NeutrinoReview with an interface for manual full text screening and adapts a legally compliant conceptual architecture from previous literature into its first practical implementation and a redesigned pipeline for automated full-text retrieval with multiple fallback mechanisms. Performance evaluation showed speedups of 8.9x for first-time PubMed downloads and 233x for subsequent retrieval from persistent local storage. By these changes, NeutrinoReview becomes an end-to-end prototype for efficient systematic reviews, combining the potential of LLMs and human-in-the-loop screening.

Supervisor: Elias Sandner, Andreas Wagner

00:00:00 Slide 1

00:00:14 Slide 2

00:00:41 Slide 3

00:01:38 Slide 4

00:01:57 Slide 5

00:02:59 Slide 6

00:03:10 Slide 7

00:03:33 Slide 8

00:03:53 Slide 9