Ansari: A Retrieval-Grounded Islamic AI Assistant -- Architecture, Deployment, and Lessons from 140,000 Conversations

The research paper introduces Ansari, a retrieval-grounded Islamic AI assistant designed to address the risks of factual fabrication and value misalignment inherent in general-purpose large language models (LLMs) when answering religious questions. Ansari utilizes an agentic retrieval loop, querying authenticated Islamic corpora to ensure responses are based solely on verifiable sources, with citations provided. Since its deployment in June 2023, Ansari has managed over 140,000 conversations across more than 25 languages.

Why it matters

This development is strategically important as it demonstrates a practical solution for deploying AI systems in sensitive domains where factual accuracy and adherence to specific value frameworks are paramount. It mitigates significant risks associated with unconstrained generative AI, offering a model for ensuring reliability and trustworthiness in AI applications across various cultural and specialized knowledge areas.

Key insights

• General-purpose LLMs pose significant risks of factual fabrication and subtle value misalignment when handling religious content, specifically for Islamic queries.
• Ansari is a deployed, retrieval-grounded Islamic AI assistant addressing these risks through an agentic retrieval loop.
• The system queries authenticated Islamic corpora, including the Qur'an, hadith collections, a multi-volume jurisprudence encyclopedia, and exegetical sources.
• Ansari's operational principle is to answer questions exclusively based on retrieved information, with citations for verification.
• Since June 2023, Ansari has facilitated over 140,000 conversations in more than 25 languages, demonstrating practical application and user engagement.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.20390