Antony Seabra, Claudio Cavalcante and Sergio Lifschitz, PUC-Rio - Pontifical Catholic University of Rio de Janeiro, Brazil

This study explores techniques for retrieving data from documents, knowledge graphs, and databases using Large Language Models (LLMs), specifically leveraging OpenAI’s GPT models as foundational frameworks for embeddings and conversational models in question-answering (QA) systems. Our research focuses on the utilization of Prompt Engineering, Retrieval-Augmented Generation (RAG), and Text-to-SQL techniques to effectively extract information from these diverse data sources without the need for model retraining. A key aspect of our study is the emphasis on explainability, demonstrating how these techniques can reveal the rationale behind retrieved information and enhance the understanding of results. We highlight the challenges encountered in specific use cases during our tests and present effective strategies and solutions to overcome them. Our findings demonstrate the potential of LLMs to surpass traditional search and retrieval systems, paving the way for more efficient and comprehensible information systems.

Information Retrieval, AI, Explainability, Documents, Knowledge Graphs, Databases, Recommendation System