Abstract
This diploma thesis presents the design and implementation of a specialized chatbot for web
platforms and describes and analyzes the services available at the local level. The approach
relies on the use of large language models (LLMs) and mechanisms for knowledge retrieval
and training, with the goal of producing documented answers that draw content from the web
platform hosting the system. The study examines the concept of the chatbot, the ways in which
training of the language model is conducted, as well as the ROCm platform used to accelerate
computation. In the context of the experimental evaluation, multiple language models were
tested using core criteria: answer accuracy, response time, and system resource requirements at
the consumer level. The proposed system is organized in architecture comprising a web
interface for user interaction, a preprocessing pipeline, and a large language-generation module
that composes the final answer with explicit citations. By combining live updating, language
generation and optimal user interaction, the chatbot aims to facilitate users in searching for
information and resolving questions based on the files of the web platform on which it operates.
Keywords:
Chatbot, conversational systems, language models (LLMs), web platforms,
artificial intelligence, Python.