Conception d’un Outil de BI Conversationnelle Basé sur le RAG Agentic

Abstract

This thesis presents the design and development of an intelligent analytical assistant system based on a multi-agent Agentic RAG architecture. The system enables users to query heterogeneous data sources using natural language. It incorporates semantic schema retrieval mechanisms and a modu- lar reasoning process to automatically generate queries, produce contextualized analyses, and create adapted visualizations. The system was evaluated across four dimensions: schema retrieval quality, query generation ac- curacy, reasoning coherence, and response relevance. The results demonstrate the effectiveness of the Agentic RAG approach for conversational Business Intelligence systems, improving flexibility, adaptability, and overall user experience.