We present Muslim, a production Arabic voice AI platform serving grounded, sourced Islamic knowledge to real users. Beyond a real-time voice pipeline (NeMo Arabic ASR, an OpenAI-compatible LLM endpoint, self-hosted TTS) and a deterministic multi-source retrieval layer routed across six Model Context Protocol servers, we report three things a research prototype typically lacks. First, a released family of fine-tuned Arabic Islamic model artifacts: an efficient tool-routing LLM (Muslim-6B-PRO, 5.94B parameters) and a Modern Standard Arabic TTS model (Fasih-TTS-V1) that ranks 5th of 17 overall and 2nd of 11 open-weight systems on the community-voted Arabic TTS Arena for MSA. Second, an account and metering layer - a free per-account turn allowance, capacity-aware refusal, and email verification deferred to the point it actually matters - that turns an open demo into an operable, abuse-resistant product. Third, a three-layer observability stack (liveness, error reporting, product analytics) built specifically around the system's characteristic failure mode: a GPU-bound agent host going silent while the web tier keeps serving normally. We report real, measured latency and accuracy figures (98.4% recitation-validation accuracy on 124 cases; end-to-end voice latency of 0.9-1.7s) and discuss the concrete engineering trade-offs and limitations of running an Islamic-knowledge voice product in production.
Figures & tables
Figure 1: System overview. The browser connects over WebRTC to LiveKit; the stateless agent runs the voice pipeline and dispatches tool calls to six MCP servers (Table 1 ) – three operator-controlled, three external.
Server
Locus
Role
Validator
local
Recitation validation (§ 7 )
IslamicMCP
local
Tafsir lookup, audio URLs
Hadith server
local
50k+ narrations, 17 collections
Qur’anic metadata
remote
Revelation context, qira’at, morphology
Web search
remote
Contemporary topics, neural + keyword
Fiqh Q&A
remote
Scholarly Q&A corpus (optional)
Table 1: The six MCP servers the agent can call. Three run on operator-controlled infrastructure; three are external services reached over HTTPS with no operator data retained.
Model
Summary
Muslim- 6B-PRO
QLoRA fine-tune (4-bit NF4, r=16 ) Dettmers et al. (2023) ; Hu et al. (2022) of a 5.94B-parameter Qwen3-architecture base Qwen Team (2025) , 262K context, on 2,731 curated tool-calling / production-derived examples via TRL von Werra et al. (2020) . Built for tool-call routing (31 tools) and persona discipline, not unsupported scripture recitation. Apache-2.0 NightPrince (2026b) .
Fasih- TTS-V1
Fine-tune of Coqui XTTS v2 Casanova et al. (2024) on 1,297 curated single-speaker MSA clips ( ∼ 2.4h), with a custom Arabic front-end (normalization, diacritization, chunking). CER 1.3% (vs. 1.8% human-recording baseline) on the SILMA Open-Source Arabic TTS Benchmark SILMA AI (2026) , ranking highest for intelligibility among open-source systems tested. Coqui Public Model License (non-commercial) NightPrince (2026a) .
Table 2: The two primary released model artifacts. Both are used in the production pipeline behind an environment-switchable configuration; the LLM currently serving live conversation traffic is a larger general-purpose model reached through the same OpenAI-compatible interface, with Muslim-6B-PRO as the path toward fully self-hosted inference.
Large language models can generate fluent responses to Islamic questions while introducing factual errors that are difficult to identify. This paper presents our system for \textsc{HalluScoring 2026} Task 2.1, \textit{Islamic Hallucination Detection and Find the Truth}. The task requires a unified two-step prediction: determining whether an Arabic answer generated by an LLM is hallucinated and selecting the verified answer from six closely related candidate options. We use the Islamic knowledge dataset provided by the shared task, which contains 600 question--answer instances, including 341 hallucinated and 259 non-hallucinated answers. Our system is based on the fine-tuned \texttt{google/gemma-4-12B-it} model and uses deterministic decoding during inference. The generated outputs are normalized to extract the hallucination label and the selected option. The system achieves a Macro-F1 score of 0.928 and a label accuracy of 0.935 for hallucination detection, together with an option accuracy of 0.895 for answer selection. These results yield a combined score of 0.912, demonstrating strong performance across both stages of the task. The lower option-selection accuracy indicates that distinguishing the verified answer from plausible alternatives remains more challenging than detecting hallucinated responses.
Large language models (LLMs) are increasingly used for knowledge-intensive question answering, including religious and legal questions. Islamic knowledge is a particularly demanding setting: answers are expected to be grounded in authoritative sources, citations must be exact, Arabic varieties differ substantially from the language of classical sources, and legitimate jurisprudential disagreement must be represented rather than collapsed into a single answer. This survey reviews the emerging field of Islamic LLMs and trustworthy Islamic AI. We organize the literature around Arabic NLP and Arabic-centric LLMs, Islamic NLP resources, Qur'anic question answering, Islamic knowledge benchmarks, retrieval-augmented generation, Islamic legal reasoning, inheritance reasoning, hallucination evaluation, and trustworthiness. We argue that fluency in Arabic is not sufficient for Islamic AI. Reliable systems require curated sources, retrieval and verification modules, citation-aware generation, madhhab-aware reasoning, human expert evaluation, and benchmarks that measure not only answer accuracy but also faithfulness, source validity, and reasoning quality. The survey concludes with a research agenda for hallucination-resistant Islamic AI systems.
Large Language Models (LLMs) voice assistants are commonly built as cascaded Automatic Speech recognition (ASR) to LLM systems, where recognition errors can distort user intent. Dislikes may also arise from ambiguous, out-of-domain, or non-request turns, making it hard to isolate ASR effects. We release WASIL (it denotes connection or linking in Arabic): in-the-wild Arabic spoken interaction prompts with audio, ASR hypotheses, assistant responses, and explicit like/dislike feedback (8,529 turns; 14.2% dislikes), plus a 2,000-turn test set covering Modern Standard Arabic (MSA) and four major dialects with their labels. We provide low-cost gold transcripts via multi-ASR agreement-guided post-editing and annotate answerability (answerable, ambiguous/needs-clarification, unsupported, not-a-request/noise) to separate intrinsic unanswerability from ASR-induced degradation. Finally, we describe scalable reference-free evaluation of responses from ASR vs. gold transcripts using multi-judge LLM scoring.