HQDR and HQPS: A Holy Quran digital repository with advanced pattern search and AI-enhanced natural language interface

Abstract

Abstract Existing digital Quran repositories support verse-level access, but few preserve the complete chapter-to-character decomposition and the full grapheme inventory of Mushaf Al-Madinah within a single integrated relational schema. We address this gap with the Holy Quran Digital Repository (HQDR), which spans chapters to characters and retains all 84 graphemes of the Hafs ‘an ‘Asim edition, including Uthmanic and Tajweed marks. Built on HQDR, the Holy Quran Pattern Search (HQPS) engine supports boundary-aware regular-expression queries directly over this representation. To make these queries accessible to non-technical users, a Model Context Protocol layer lets Large Language Models translate natural-language requests into expert-validated patterns that HQPS executes deterministically. We evaluate the system along two complementary axes. On a keyword-retrieval benchmark, HQPS achieves exact-match accuracy on every query, against 70% for Tanzil and 20–40% verse-level accuracy for KFGQPC and Alfanous (2024). As a domain-specific correctness validation—rather than a general information-retrieval benchmark—exhaustive extraction across five Tajweed rule families returns 45,460 instances, validated by three certified experts with 100% precision on four families and 99.98% on Lam Qamariyyah.