Natural Language Interaction for a BI Platform and Data Integration:Task-Oriented, Retrieval-Augmented Agent Using LLMs

This thesis focuses on the development of a chatbot designed to facilitate interaction with a BI platform using natural language commands. The primary objective was to create a system capable of interpreting user requests and translating them into database operations, thereby simplifying data access and management. The project utilized the LangChain library in Python to build an agent capable of executing queries and commands on a database. An incremental development approach was adopted, starting with a foundational model and progressively enhancing it with advanced functionalities to align with the structure of Statwolf's status queries. The chatbot was developed during an internship at Statwolf, specifically tailored to interact with their datasources. It was designed to transform natural language queries into a structured dictionary-like format compatible with Statwolf's data query system. Additional functionalities were incorporated to support the creation of new metrics and dimensions through natural language commands, which were then translated into ClickHouse query definitions.