A Multi-Purpose Text Generation Framework Using LLM: OWN-GPT

Authors: Kinal Patel, Rauki Yadav, Payal Naykawala, Bhumi Torani

DOI Link: https://doi.org/10.22214/ijraset.2026.78760

Abstract

OWN-GPT is a 2.3-billion-parameter transformer-based Large Language Model (LLM) developed as a full-stack AI system featuring a React.js frontend and a Python FastAPI backend. The system provides conversational text generation, image-based question answering using OCR and visual feature extraction, and PDF document question answering through Retrieval-Augmented Generation (RAG). The backend incorporates PyTorch, HuggingFace Transformers, PyMuPDF, LangChain, FAISS, sentence-transformers, OpenCV, Tesseract, and ViT-Base/16 for multimodal processing. Trained on an 81 GB multi-domain corpus (~40 billion tokens), OWN-GPT achieves a perplexity of 7.8, BLEU score of 0.68, image module CER of 6.8%, and RAG module faithfulness and relevancy scores of 0.81 and 0.79. Designed for local deployment in Indian academic institutions, the system ensures data privacy and zero per-query operational cost. A user study reports a System Usability Scale (SUS) score of 77.1, confirming strong usability and user satisfaction.

Introduction

OWN-GPT is a locally deployable, multimodal Large Language Model (LLM) designed for academic institutions to ensure data privacy, avoid cloud costs, and provide on-premise AI capabilities. It integrates conversational AI, image question answering via OCR and Vision Transformers, and PDF-based Retrieval-Augmented Generation (RAG). The system uses a FastAPI backend, React.js frontend, and PyTorch-based transformer models trained on an 81?GB multi-domain corpus. Evaluation shows strong language generation, OCR-integrated image understanding, and accurate PDF retrieval, with high user satisfaction, demonstrating an efficient, privacy-preserving, and practical AI platform for institutional deployment.

Conclusion

OWN-GPT demonstrates that a fully local, multimodal LLM system can be successfully developed and deployed within academic institutions. Its strong performance across text generation, OCR-based image understanding, and PDF-RAG-based document analysis makes it a valuable tool for teaching, research, and administration. The system’s cost-free operation post-deployment addresses major barriers faced by educational organizations.

References

[1] V. Visweswaraiah, \"Everyday AI: Real-World Applications of Transformer-Based Large Language Models,\" IJCTT, vol. 73, no. 9, pp. 19-27, 2025. [Online]: https://ijcttjournal.org [2] A. Singh et al., \"Research Gaps in Developing Fair and Inclusive LLMs for India\'s Agriculture Sector,\" IRJET, vol. 12, no. 11, pp. 550-556, 2025. [3] M. A. Khan, R. Sharma, S. Patel, \"Large Language Models: An Overview of Foundational Architectures, Recent Trends, and a New Taxonomy,\" Discover Applied Sciences, Springer Nature, doi:10.1007/s42452-025-07668-w, 2025. [4] G. Yenduri et al., \"GPT (Generative Pre-Trained Transformer) — A Comprehensive Review on Enabling Technologies, Potential Applications, Emerging Challenges, and Future Directions,\" IEEE Access, vol. 12, pp. 54608-54649, doi:10.1109/ACCESS.2024.3387368, 2024. [5] S. Suresh, M. Chandrika, \"Large Language Model (LLM): An AI Model for Pattern Recognition,\" IJERT, 2023. [6] A. Gupta, R. Rao, S. Mehta, \"Advancements in Transformer Architectures for Large Language Models: From BERT to GPT-3 and Beyond,\" IRJMETS, vol. 6, no. 4, 2024. [7] T. B. Brown et al., \"Language Models are Few-Shot Learners,\" Advances in NeurIPS, vol. 33, pp. 1877-1901, 2020. [8] A. Vaswani et al., \"Attention Is All You Need,\" NeurIPS, vol. 30, arXiv:1706.03762, 2017. [9] H. Touvron et al., \"LLaMA 2: Open Foundation and Fine-Tuned Chat Models,\" arXiv:2307.09288, 2023. [10] S. Biderman et al., \"Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling,\" ICML 2023, arXiv:2304.01373. [11] J. Devlin et al., \"BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding,\" NAACL-HLT 2019, pp. 4171-4186, doi:10.18653/v1/N19-1423. [12] C. Raffel et al., \"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer (T5),\" JMLR, vol. 21, no. 140, 2020. [13] A. Radford et al., \"Language Models are Unsupervised Multitask Learners (GPT-2),\" OpenAI Blog, 2019. [14] Z. T. Hamad et al., \"ChatGPT\'s Impact on Education and Healthcare,\" IEEE Access, vol. 12, pp. 114858-114876, doi:10.1109/ACCESS.2024.3437374, 2024. [15] P. Liu et al., \"Pre-Train, Prompt, and Predict: A Systematic Survey of Prompting Methods in NLP,\" ACM CSUR, vol. 55, no. 9, doi:10.1145/3560815, 2023. [16] M. Ansari et al., \"Intelligent Chatbot,\" IJERT, vol. 10, no. 3, 2021. [17] A. Binu et al., \"Chatbot Using Artificial Intelligence,\" IJERT, vol. 12, no. 6, 2023. [18] M. Senthil Kumar et al., \"An Automated Chatbot for an Educational Institution using NLP,\" IJCRT, vol. 10, no. 5, pp. 142-150, 2022.

Copyright

Copyright © 2026 Kinal Patel, Rauki Yadav, Payal Naykawala, Bhumi Torani. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Download Paper

Paper Id : IJRASET78760

Publish Date : 2026-03-25

ISSN : 2321-9653

Publisher Name : IJRASET

DOI Link : Click Here