Document Query Processing With Mistral 7B and Streamlit

PDF’s form an integral part in day to day life of every individual. Researchers, scientists, teachers and students need to read and go through tons of PDF in order to gain information. This is an overhead for the people. In past there have been multiple attempts to get the information from the PDF’s by directly asking the question. But most of the approaches are limited to single PDF and the accuracy is a concern with respect to the context in which answer is generated. In this paper we propose an approach that used Mistral 7B model to generate the answers to the questions asked by the user on multiple PDF’s uploaded at once. We also evaluate our model against the existing approaches to answer generation and demonstrate a better approach to solve the problem. Upon analysis we arrive at a conclusion that Mistral-7B outperforms other LLMs for this task. By incorporating prompt lookup technique, the response time was improved by approximately 80%.