Development of an Application with Virtual Assistant Based on LLM for the Knowledge Domain on Anemia

This study aimed to develop an intelligent virtual assistant for anemia management using Retrieval-Augmented Generation (RAG) integrated with large language models (LLMs). Through a comparative experimental methodology, four models — GPT-4 Turbo, LLaMA-3-70B, Mistral 7B, and Gemini 1.5 Pro — were evaluated before and after RAG implementation. The results showed notable improvements, with LLaMA-3-70B reaching 89% diagnostic accuracy (a 17% increase) and an F1-score of 0.85 in anemia subtype classification. The system achieved 95% accuracy in nutritional anemia and 93% accuracy in emergency cases, while maintaining 82% accuracy in complex cases. User testing indicated high satisfaction (4.6/5) and strong clinical utility. In conclusion, integrating RAG-based retrieval, medical domain expertise, and mobile accessibility enhanced diagnostic precision and knowledge access, demonstrating the potential of specialized AI to improve anemia management in resource-limited healthcare environments.