DUGi

Tuning a generative language model for the creation of customized chatbots for Public administrations

http://dugi.udg.edu/item/http:@@@@hdl.handle.net@@10256@@27576

This master’s thesis explores the development of a customized chatbot system using GPT (Generative Pre-trained Transformer) and RAG (Retrieval Augmented Generation) technologies to improve the efficiency of public administration services in Catalonia, specifically targeting social rights and benefits. The project, conducted in collaboration with DXC Technology, focuses on creating a multilingual, accessible, and user-friendly interface while optimizing the chatbot’s performance in terms of accuracy and cost. Key components of the system include: • Frontend Interface: Built with Angular to handle user interactions. • Backend System: Manages the conversation flow, connects with Azure services, and employs a Vector Search API for information retrieval. • Data Preparation: Involves sourcing information from public websites and processing it to improve chatbot responses.

9