Deen Guide: Verifying, Not Trusting, LLM Answers in Islamic Question Answering

Large language models (LLMs) provide fluent answers to religious questions, but fluency can obscure fabricated citations, uncertain rulings, and inappropriate confidence. These risks are particularly important in Islamic question answering, where users may interpret model responses as religious guidance. We present Deen Guide, a working source-grounded Islamic question-answering assistant designed for Muslim community events. Rather than relying on prompting alone, Deen Guide combines a small human-reviewable Islamic knowledge corpus, restricted-domain web search, code-level citation verification, and a three-tier trust model that labels every response as Verified, Not Verified, or Declined. Verified answers must be grounded in curated Qur'anic or authenticated hadith sources, while questions beyond the reviewed corpus are explicitly marked as unverified or redirected to a human scholar when individualized or contested guidance is required. The prototype has been deployed at a community event and includes speech interaction, scholar-review functionality, feedback logging, and an audit dashboard. Preliminary internal testing demonstrates the feasibility of code-enforced grounding while also revealing instability in classification for some borderline questions. Deen Guide illustrates a practical design pattern for building safer community-facing AI systems in sensitive religious knowledge domains.