Research paper AI assistant using retrieval augmented generation and multimodal LLM

Mohd Rayhan, Ammar Helmey Iskandar and Kamaru-Zaman, Ezzatul Akmal and Aidil Rozaidi, Muhammad Fareed and Ahmad, Azlin (2025) Research paper AI assistant using retrieval augmented generation and multimodal LLM. In: E-proceedings of international tinker innovation & entrepreneurship challenge (i-TIEC 2025). International Tinker Innovation & Entrepreneurship Challenge (2nd). Universiti Teknologi MARA Cawangan Johor Kampus Pasir Gudang, Universiti Teknologi MARA, Johor, pp. 248-262. ISBN 978-967-0033-34-1

Abstract

Researchers often face challenges such as the time-consuming process of document review and the complexity of comprehending technical concepts during journal writing. Emerging large language models, such as ChatGPT and Poe, have proven to be valuable tools for streamlining the writing process. However, these models often require payment for extensive usage or document uploads, posing accessibility barriers. This innovation is motivated by the need to simplify research workflows and enhance the understanding of research documents while ensuring these capabilities are accessible to university users without incurring additional costs. The Research Paper AI Assistant is an innovative system designed to revolutionize how researchers interact with and synthesize information from academic papers. Leveraging advanced technologies such as the Large Language Model (LLM) - Gemma 9b, Retrieval-Augmented Generation (RAG), and the Chroma Embedding Multimodal Model, this innovative system features an intuitive chatbot interface. It enables users to upload multiple PDFs and conduct conversational interactions to extract and synthesize complex information. The scope of the system is specifically confined to the context of the uploaded documents, ensuring focused and relevant responses. The system features automatic diagram analysis for detailed insights and a dedicated "Future Work" tab that instantly identifies key problems and future research directions outlined by authors subsequently, promotes efficient knowledge acquisition. Its unique ability to generate notes, explain complex terms, and provide instant retrieval of relevant content makes it invaluable for young academics such as students and professionals. The system’s socio-economic impact extends beyond accelerating research breakthroughs and fostering innovation in academic insights. It also significantly benefits students by simplifying the process of understanding complex academic materials, aiding in literature reviews, and enhancing their research and learning experiences. By providing an accessible and interactive tool, students can efficiently extract relevant information from academic papers, develop critical thinking skills, and improve their academic writing. Commercially, the system holds immense potential for integration into academic institutions, research labs, and publishing platforms, supporting applications ranging from education to corporate R&D.

Metadata

Item Type: Book Section
Creators:
Creators
Email / ID Num.
Mohd Rayhan, Ammar Helmey Iskandar
UNSPECIFIED
Kamaru-Zaman, Ezzatul Akmal
UNSPECIFIED
Aidil Rozaidi, Muhammad Fareed
UNSPECIFIED
Ahmad, Azlin
UNSPECIFIED
Contributors:
Contribution
Name
Email / ID Num.
Advisor
Zainodin @ Zainuddin, Aznilinda
314217
Subjects: L Education > LB Theory and practice of education > Teaching (Principles and practice) > Technology. Educational technology
L Education > LB Theory and practice of education > Educational technology
L Education > LB Theory and practice of education > Teaching (Principles and practice) > Teaching aids and devices
Divisions: Universiti Teknologi MARA, Johor > Pasir Gudang Campus > College of Engineering
Series Name: International Tinker Innovation & Entrepreneurship Challenge
Number: 2nd
Page Range: pp. 248-262
Keywords: Large language model, Retrieval-augmented generation, Diagram analysis, Future work extraction, Knowledge accessibility
Date: 2025
URI: https://ir.uitm.edu.my/id/eprint/119095
Edit Item
Edit Item

Download

[thumbnail of 119095.pdf] Text
119095.pdf

Download (8MB)

ID Number

119095

Indexing

Statistic

Statistic details