A web system for LLM-based Q&A

• Main Author: Banerjee, Tanya
• Other Authors: Long Cheng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Q&A
• Online Access: https://hdl.handle.net/10356/175236
• Institution: Nanyang Technological University

Chatbots based on Large Language Models have gained popularity in the last year. However, they may not be informed on all topics, especially recent events or a user’s queries about very specific subjects. These LLMs have token limits so the user cannot copy large amounts of text into the chat interface. There is also a growing number of chatbots available for users to choose from, and they have to explore each option before deciding on one that suits their needs. Therefore, it would be helpful to have a system where the user can provide some text data (may be larger than token limit) and then ask questions about the topics in the text. The context sent to the LLM should be concise and relevant to the questions so as not to exceed the token limit. Given such a context, an LLM would be able to answer the required questions. The user should also be able to choose between a few LLM options so that they can compare all of them in one place. The aim of this project is to develop a web system that allows users to input text in three ways: uploading a text file, providing a link to a web page and pasting text into an input component. They can choose an LLM from the provided options and ask questions about the text using a chat interface. The Q&A will be implemented using Retrieval Augmented Generation.