Muslim: A Deployed Arabic Voice AI Platform for Grounded Islamic Knowledge

Abstract

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.

In numbers

End-to-end voice latency

0.9–1.7 s

Recitation validation (122 / 124)

98.4%

Open-weight TTS, Arabic TTS Arena (MSA)

2nd of 11

Parameters, Muslim-6B-PRO

5.94B

Contributions

1. A real-time Arabic voice pipeline integrating Arabic-specialized ASR, an environment-configurable LLM endpoint and self-hosted TTS, with a six-server tool-retrieval layer.
2. A released family of fine-tuned Arabic Islamic model artifacts, including an MSA TTS model independently ranked on a community leaderboard.
3. An account and metering layer — a free rolling turn allowance, a fail-open capacity wall, deferred email verification — undocumented in prior Islamic voice AI work, because prior work did not need to survive real traffic.
4. A three-layer observability design aimed at a specific, non-obvious failure mode of deployed systems.
5. An evaluation that combines real measured latency and accuracy with reliability evidence from an automated test suite.

Released artifacts

• ModelMuslim-6B-PROThe tool-routing language model that answers users in production.
• ModelMuslim-6B-PRO-GGUFQuantized build of the same model for local inference.
• ModelFasih-TTS-V1The self-hosted Modern Standard Arabic voice.
• DatasetFasih-TTS-BenchmarkEvaluation set for the voice.

Cite

Yahya Mohamed Elnawasany. Muslim: A Deployed Arabic Voice AI Platform for Grounded Islamic Knowledge. arXiv:2609.31511 [cs.CL], 2026. doi.org/10.48550/arXiv.2609.31511

@misc{elnawasany2026muslim,
  title         = {Muslim: A Deployed Arabic Voice AI Platform for
                   Grounded Islamic Knowledge},
  author        = {Elnawasany, Yahya Mohamed},
  year          = {2026},
  eprint        = {2609.31511},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  doi           = {10.48550/arXiv.2609.31511},
  url           = {https://arxiv.org/abs/2609.31511}
}