Hikmah AI: Developing a Specialized Large Language Model for Islamic Knowledge and Information Retrieval

Abstract

This research aims to develop an Islamic-based Large Language Model (LLM) system called Hikmah AI, designed to provide valid, contextual Islamic information sourced from the Qur'an, Hadith, and trusted Islamic literature. The background of this research is driven by the increasing public demand for Islamic knowledge in the digital era, while many available sources remain unstructured and lack credibility. The system was developed using the OpenAI GPT-5 model integrated with the Retrieval-Augmented Generation (RAG) method based on hybrid search (vector and keyword). The development followed the Waterfall methodology, encompassing the stages of requirements analysis, design, implementation, testing, and maintenance. The testing process employed evaluation metrics consisting of Precision, Recall, F1-Score, Relevance, and Accuracy to assess the quality and relevance of the model's responses to Islamic queries. The test results demonstrate that the system successfully achieved Precision, Recall, and F1-Score of 100% across all test queries, along with an average Relevance and Accuracy of 86.2%. Expert validation by an Islamic scholar concluded that the system is highly feasible, with citations consistent with the original scriptural texts and no substantive errors found. The results of this research demonstrate that Hikmah AI has successfully become an innovative solution for providing accurate, accessible Islamic information in accordance with authentic Islamic sources.