DocBot Connect: A Conversational Search System Powered by Large Language Models
Ratna Nitin Patil, Sandesh Ashok Buchkul, Ritesh Sanjay Mohite, Jayesh Samadhan Borase, Anuj Khandelwal, Shitalkumar Rawandale · 2024
DocBot Connect revolutionizes document retrieval by enabling users to have a natural conversation to search through multiple PDFs. This user-centric system leverages the power of large language models (LLMs) and cutting-edge information retrieval techniques.DocBot Connect boasts a user-friendly interface where users can upload their PDFs and ask questions in plain English. To unlock the deeper meaning within these documents, the system utilizes Google PaLM embeddings. These embeddings capture the essence of the text, allowing for a richer understanding. A critical step involves text segmentation, which meticulously divides the document content into manageable chunks. These segmented portions are then processed by Google PaLM, resulting in a robust embedded representation. This embedded representation forms the foundation for efficient retrieval using a FAISS vector store, enabling lightning-fast information retrieval. To ensure a seamless and contextually aware conversation, DocBot Connect incorporates a ConversationBufferMemory component. This component meticulously stores the conversation history, allowing the system to maintain a coherent dialogue flow. Finally, a ConversationalRetrievalChain, powered by the formidable capabilities of Google PaLM, retrieves relevant information from the vector store. This retrieval process is not solely driven by the user's query but also meticulously considers the established conversational context. DocBot Connect transcends traditional document retrieval systems, offering an intuitive and efficient gateway to access and comprehend information within a multitude of PDFs. This user-friendly approach fosters a more interactive and engaging experience for those navigating the ever-expanding world of document exploration.