Research
Project
D'AI: Islamic Agentic AI Framework
No.IBest Presenter — ICODS 2025
Agentic AI framework operationalizing tawhidic epistemology for Islamic knowledge retrieval with multi-layer knowledge bases (SQL + Vector + Knowledge Graph).
Impact
Built a 6,236-verse Qur'anic benchmark evaluating 15 LLMs. Agentic approach improved accuracy from 31–69% to 80–99.6% EM while reducing cost 167x. On the QRCD question-answering benchmark, prompt routing reached pAP@10 70.45%, 6.7 points above the published state of the art (63.7%).
LLMs · RAG · Knowledge Graph · SQL · Vector DB