'DB Pal' or 'PolyDatabase Guide' is a conversational AI created using large language models (LLMs) and natural language processing (NLP), designed to offer personalized educational support. The genesis of this project stemmed from my personal experiences as a first-year college student abroad during the COVID-19 pandemic. Faced with numerous challenges and pressures, I identified a critical need for a virtual assistant capable of providing immediate, accessible, and tailored educational support. While the university assigned mentors to groups of 50 students, their limited availability, especially during off-hours, often led to delayed responses and support. This experience fueled my commitment to develop a solution that could provide real-time, personalized guidance to students in need.
This AI-enhanced Learning Companion is more than a chatbot; it is a testament to the transformative power of technology in educational experiences. By offering personalized, empathetic, and continuous support, it stands as a beacon of innovation in the realm of educational technology, making a significant contribution to the academic and personal growth of students.
In this project, I undertook a thorough experimental evaluation of various large language models (LLMs) and encoder models to assess the efficiency and accuracy of the LLM pipeline. Utilizing a comprehensive set of testing criteria—such as contextual relevance of responses, grammatical integrity, effective sourcing, and computational resource demands—I ultimately chose a model that combines MPNet with Flan-Alpaca. This selection was followed by fine-tuning the pipeline to optimize response quality.