Implementation of Retrieval-Augmented Generation (RAG) and Large Language Models (LLM) for a Document and Tabular-Based Chatbot System

Rafidhul Haque, Imam Chalish

Journal of Electronics Technology Exploration (JOeTEx)
Vol. 3 No. 1 (2025): June 2025

The challenge of accessing information from disparate sources—unstructured documents and structured tabular data—hinders efficiency in enterprise information systems. This study addresses this challenge by presenting the design, implementation, and validation of a unified chatbot system powered by Retrieval-Augmented Generation (RAG) and Large Language Models (LLM). For unstructured documents, the system implements a RAG pipeline utilizing ChromaDB for vector indexing and OpenAI embeddings. Meanwhile, for structured data, it leverages a Text-to-SQL engine to translate natural language queries into SQL commands, with results visualized via QuickChart. The architecture is built upon a modular FastAPI backend with role-based access control and was rigorously validated through blackbox functional testing. Results demonstrate 100% functional success across all endpoints, confirming the architecture's reliability. This study confirms the viability of a unified RAG and Text-to-SQL architecture, offering a practical blueprint for creating more intelligent and integrated data interaction systems in enterprise environments.