Designing a Healthcare Co-Pilot with Generative AI

This paper presents our methodology for designing, testing, and evaluating a co-pilot tailored for healthcare professionals working in Spanish-speaking contexts. The co-pilot facilitates efficient access to textual information from clinical notes and structured data within Electronic Health Records (EHRs). Its primary objective is to save professionals’ time while ensuring accurate information retrieval. The system leverages a Retrieval-Augmented Generation (RAG) architecture powered by state-of-the-art Large Language Models (LLMs). Our evaluation, conducted on a dataset of 3,500 yes/no questions, achieved an F1 score of 0.82, demonstrating its effectiveness in the target domain.