What we publish about how we build: papers with DOIs, open bilingual models, and the conventions for citing our corpora in scholarly work.
How a corpus of 18,000 Arabic and Persian works is built so that no text reaches the researcher without attribution and no relation without evidence.
From the digital library to the semantic knowledge structure: the project, its place on the path of scholarly reasoning, and the disciplined role of AI within it.
Bilingual Arabic–Persian models released under the MIT licence on Hugging Face, each with a model card describing its training data, evaluation and intended use.
| Model | Task | Base | Languages |
|---|---|---|---|
| fiqh-embed-ar-fa | Bilingual embeddings for semantic search over jurisprudence (1024-d); GGUF variant available | BAAI/bge-m3 | ar · fa |
| fiqh-rerank-ar-fa | Cross-encoder reranker for the second retrieval stage | BAAI/bge-reranker-v2-m3 | ar · fa |
| fiqh-classifier-ar | Routes a question to its chapter of jurisprudence and flags vague questions | intfloat/multilingual-e5-base | ar |
| fiqh-glossary-linker | Links a term or phrase to its entry in a 2,034-term bilingual glossary | intfloat/multilingual-e5-base | ar · fa |
| turath-ocr-fix | Byte-level post-correction of OCR damage in Arabic and Persian text | google/byt5-base | ar · fa |
| sadiq-cite-8b | Retrieval-grounded citation model: quotes bearing passages verbatim with book and page; GGUF variant available | Qwen/Qwen3-8B | ar · fa |
A citation-grade research engine over the Ahl al-Bayt Library, exposed to research assistants through the Model Context Protocol. The corpus is reviewed against published standards:
Data quality: accuracy and completeness of book, author, editor and volume metadata.
Library reference model: editions of one work are kept distinct in every citation.
Sampling inspection: page attribution is recalibrated weekly on a fresh sample.
Practice of the academic Arabic-heritage corpora: author-name normalisation and duplicate prevention.
The corpus is a means of access, not a source: cite the printed book itself for every passage, and cite the corpus for the retrieval.
Sadiqoon Technologies. Moheet Corpus (version and date of query). moheet-ablibrary.net
Ali, Shady. Moheet Corpus: A Methodology for Building a Citation-Grade Semantic Corpus of Arabic and Persian Islamic Scholarship. Zenodo, 2026. https://doi.org/10.5281/zenodo.23865623
@misc{ali2026moheet,
author = {Ali, Shady},
title = {Moheet Corpus: A Methodology for Building a Citation-Grade Semantic Corpus of Arabic and Persian Islamic Scholarship},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.23865623},
url = {https://doi.org/10.5281/zenodo.23865623}
}
Sadiqoon Technologies. fiqh-embed-ar-fa (MIT). Hugging Face, 2026. https://huggingface.co/sadiqoon/fiqh-embed-ar-fa
The bibliographic knowledge graph of the Ahl al-Bayt Library and an open benchmark for entity and relation extraction from classical Arabic scholarly prose. Both will be published on Hugging Face and Zenodo.
Researchers and institutions interested in collaboration, evaluation or data partnerships are welcome to write to info@sadiqoon.uk.