Tools
Tools
Rajpoot, Abha Kiran, Virdee, Bal Singh and Khanna, Ashish (2025) A scalable framework for domain-specific knowledge enhancement in medical applications using OpenBioLLM and retrieval-augmented generation. In: International Conference on Artificial Intelligence and Networking (ICAIN-2025), 1-7 October 2025, Dubai, UAE. (In Press)
This study propose a scalable and modular framework that enhances Utilization of Large Language Models (LLMs) for specific biomedical applications via the integration of Retrieval-Augmented Generation (RAG). Employing OpenBioLLM as the foundational model, pretrained on extensive biomedical corpora with real-time retrieval from a domain-specific biomedical knowledge base. This hybrid methodology enables the model to dynamically integrate current scientific knowledge during inference, significantly improving both factual accuracy and clinical relevance across multiple benchmark datasets, including MedQA, BioASQ, PubMedQA, and MedMCQA, comparing our framework against state-of-the-art models such as GPT-4, GPT-3.5, Med-PaLM 2, BioBERT, and baseline OpenBioLLM. The proposed OpenBioLLM with RAG framework outperforms all baselines, achieving 90.0% accuracy on MedQA and substantial gains in BLEU, ROUGE-L, and BERTScore metrics. Expert evaluations further validate improvements in factual correctness (90%) and clinical relevance (88%). Our findings demonstrate that integrating retrieval mechanisms into domain-specific LLMs addresses the limitations of static knowledge and provides a highly accurate, clinically meaningful, and scalable solution for biomedical question-answering tasks.
![[thumbnail of A Scalable Framework for Domain-Specific Knowledge Enhancement.pdf]](https://repository.londonmet.ac.uk/style/images/fileicons/text.png "A Scalable Framework for Domain-Specific Knowledge Enhancement.pdf")
Restricted to Repository staff only until 15 October 2026.
Download (547kB) | Request a copy
 |
View Item |