Islamic sources, searched and recited at scholarly depth
Scholarly-grade retrieval and phoneme-level recitation
Islamic institutions, scholars and platforms need search and analysis that understands classical Arabic and religious context at the source level, and recitation tools precise enough to catch Tajweed errors a human teacher would catch.
Why general models fail on religious text
Religious material is where general-purpose models are furthest out of their depth: classical Arabic, dense jurisprudential context, and a standard of attribution that a fluent paraphrase does not meet. This is one of Kawn Lab's two standing research tracks rather than a side application. Islamic text understanding and retrieval, Quranic speech recognition, Tajweed-aware speech processing and fatwa knowledge systems are built as their own line of work.
A research track, not a feature
Islamic language technology, Quranic AI and speech processing are one of Kawn Lab's two standing research tracks, with their own models, benchmarks and published papers.
Attribution, by construction
Every generated passage maps back to its source, so the output stands up to scholarly scrutiny instead of merely sounding authoritative.
Verified data underneath
The Quranic corpus behind this is structured rather than scraped: hundreds of editions and text versions, layout coordinates matching printed Masahif, and recitations covering the major Riwayat with word-level synchronisation.
What Islamic institutions deploy
- Scholarly retrieval
The Quran, Hadith, Tafseer and fatwa collections searched as one corpus.
Search runs on Kawn Lab's Islamic embedding model, so a query is resolved by meaning and religious context rather than by keyword. The Quran, authenticated Hadith, Tafseer commentary and Islamic Q&A datasets are searchable together instead of four times over.
Islamic Knowledge AI - Recitation correction
Tajweed errors caught at the level a teacher catches them.
Nabr is an acoustic model built for Tilawa rather than a general speech model pointed at it. It analyses eight acoustic properties per sound to catch elongation, nasal resonance, letter bounce and heavy versus light pronunciation, works on live audio so feedback arrives while the reciter is still reciting, and was trained on recordings kept in their natural acoustic conditions rather than studio-clean ones.
Islamic Knowledge AI - Research drafting
A jurisprudential question turned into a sourced outline.
Tbyaan's research agent synthesises the relevant evidence, builds an outline and drafts from classical sources instead of returning a list of links. It does not replace the scholar; it removes the cross-referencing that used to consume the time analysis needed.
Islamic Knowledge AI
Already running in the sector
A pilgrim services platform
Accurate Arabic speech recognition and dialect-aware translation were needed to guide pilgrims at scale.
Misraj's Arabic speech and translation research powers multilingual guidance without losing dialectal nuance.
Guidance now runs across multiple languages while preserving dialectal accuracy at scale.
Deployment options
Cloud API access for platforms that want to build quickly, or on-premises deployment where the corpus and its usage have to stay inside the institution. The Trust Center sets out the deployment options in full.
Solutions
Let's talk about what you're trying to build.
Tell us the problem. We'll tell you honestly whether AI is the right answer.