Desarrollo de un Chat Bot para el laboratorio clínico

Abstract

The article describes the development of an AI-based chatbot designed to reduce errors in the pre-analytical phase of the laboratory through 24/7 responses based on a validated catalog of about 4,000 tests, converted to JSON and using the RAG (Retrieval Augmented Generation) technique on GPT-4o-mini. It has been implemented in Python using libraries such as LangChain and FAISS and features a web interface to improve accessibility. The system also includes a memory server to speed up responses, and the bot integrates with databases containing images of tubes and extraction materials. It can be classified as Class I Medical Device Software. Pilot tests are being conducted in two laboratories where 80% of users find it useful, although 30% encounter difficulties in formulating queries, highlighting the need to improve user experience and automatic language detection. RAG has helped minimize "hallucinations" and update content without retraining, but technical (heterogeneous table handling), regulatory (GDPR), and clinical explainability challenges persist. The tool is expected to be expanded to the analytical and post-analytical phases and potentially evolve towards clinical decision support functions under CE regulations, thus enhancing the quality of the clinical laboratory and freeing up staff for higher-value tasks.

Date of Award	Jul 2025
Original language	Spanish
Supervisor	Remo Suppi Boldrito (Director)