Design and Implementation of a Multilingual Conversational Chatbot for the HR Department

In the current business scenario, the effective management of human resources is a critical aspect. Organizations are implementing new technologies such as artificial intelligence (AI) to transform HR operations, increase efficiency, and engage employees. Of all these AI applications, chatbots are particularly useful in automating processes, offering quick assistance, and personalization. With the help of technologies like NLP and machine learning, HR chatbots are capable of answering a number of questions and tasks, which include answering general queries, arranging interviews, onboarding new employees, and managing employee feedback. This automation enables HR professionals to spend more time on strategic work, thus enhance the performance of the HR department. The Retrieval-Augmented Generation (RAG) is a creative solution to construct chatbots that leverage the best from both worlds – retrieval-based and generation-based models to obtain accurate and contextually suitable answers. In RAG when a user asks question the system will then extract from an existing database the most related data. This retrieved context is then passed on to an LLM to provide a detailed and well-structured response. In this way, with the help of both the retrieval and the generation, RAG makes sure that the answers given by the chatbot are not only helpful but also contextually relevant and improving the experience. This thesis focuses on the development of an HR chatbot for a multinational company using the Retrieval-Augmented Generation (RAG) model. RAG improves the chatbot’s response generation by incorporating retrieval-based and generation-based methods. To achieve the best results, three RAG models were implemented and compared. Every approach was designed to enhance the chatbot’s ability to comprehend and generate responses. For each user query, the most relevant dataset segments were identified and then fed into large language models (LLMs) to produce detailed answers. The chatbot also had a conversational history to make sure that the conversation is logical and relevant to the current context and it also able to understand and respond in Italian. Comparing the three different RAG implementations helped in determining the best method to apply in the HR chatbot. To compare the performance of the different implementations, the RAGAS (Retrieval-Augmented Generation and Answer Synthesis) framework was used. The evaluation considered the retrieval context relevance and the quality of the generated responses in terms of faithfulness and answer relevance. Based on this evaluation, the advantages and disadvantages of each approach were identified to determine the best solution for the HR chatbot. This study contributes to the understanding of how the use of AI-driven chatbots has the possibility of enhancing the HR practices through efficiency, accuracy, and employee satisfaction. When properly designed and integrated, an HR chatbot can help free up time for HR specialists, enabling them to concentrate on key business processes that enhance the performance of the organization.