Implementazione e validazione di una pipeline RAG-based per l’automazione delle indagini epidemiologiche: analisi di performance e applicabilità nel contesto dei sistemi di sorveglianza sanitaria italiana.

Abstract

We implemented a three-phase AI pipeline for automating epidemiological investigations: structured PDF data extraction, RAG-driven report generation, and final document assembly. Expert validation (n=200) yielded high scores for completeness (4.7), accuracy (4.5), relevance (4.6), clarity (4.8), and timeliness (4.4), with inter-rater κ=0.85 and a 60% time reduction. The system is scalable to serve other surveillance systems.