uMentor: LLM-Powered Chatbot for Harnessing Technology Books in Digital Library

Large language models (LLMs) are currently attracting considerable interest and being extensively utilised across all disciplines. Chatbots such as ChatGPT are used by many people and are also employed as internal personal assistants in numerous firms. However, in education, ChatGPT is currently limited in providing comprehensive knowledge and resources related to technology curriculum, with a most focus on English material. To some extent, the access, exploitation and comprehension of the content and information in technological fields in terms of materials and knowledge are relatively limited for our local Vietnamese students due to individual skills and linguistic obstacles. The study presents uMentor, a method that uses the pretrained LLaMA-2 model trained on Vietnamese language datasets to help students enhance learning quality in technological fields with the aid of a virtual mentor. For uMentor, we firstly propose a process to collect, normalize, and improve a specific instruction dataset on technology themes in Vietnamese. We then perform the fine-tuning the pretrained LLaMA-2 model with our instruction dataset. Consequently, we build a virtual mentor prototype to assist students with academic issues and offer assistance on technology book discovery, study road-maps, schedules, and technology-related resources. The preliminary results indicate a notable enhancement in the learning and teaching environment because of uMentor’s efficacy in resolving knowledge deficiencies and increasing academic engagement.