Islamic Knowledge AI
Islamic knowledge, retrieved and analysed at the scholarly-source level
General-purpose models miss the classical Arabic, the religious context and the phoneme.
Islamic research and recitation tools built on general-purpose models miss classical Arabic nuance, religious context and the phoneme-level precision that Tajweed requires.
Delivered by
Qurani.ai pulls together Quranic text, audio and translation into one API, with semantic search that outperforms Gemini, Voyage and Cohere on Islamic retrieval at 0.83 average. Tbyaan turns a jurisprudential or historical question into a sourced, structured draft. Our speech research also produced Nabr - a Quranic recitation model analysing 8 acoustic properties per sound at 0.59% character error rate - the strongest evidence we build speech models at the phoneme level rather than fine-tuning someone else's.
Why this needs its own models
An embedding model trained on the corpus
Kawn-Embed-Islamic is trained on Islamic Q&A, Tafseer commentary and Hadith records, so retrieval captures dense religious context and classical Arabic semantics instead of matching keywords.
Tajweed is an acoustic problem, not a text one
Nabr analyses eight acoustic properties per sound: elongation timing, nasal resonance, letter bounce, heavy and light pronunciation. These are the things generic voice models miss entirely.
Sourced output, not authoritative-sounding output
Tbyaan maps every piece of generated text back to its source, so a draft holds up to scholarly scrutiny rather than merely reading as though it would.
What it gives a builder and a researcher
Quranic text, audio and translation in one interface
360+ Quran editions and 270 text versions with Uthmanic script, multiple Qira'at and layout coordinates matching printed Masahif.
Recitation audio with word-level synchronisation
354,000+ audio files across 34 reciters covering the major Riwayat, with word-level sync for highlighted playback.
Real-time recitation correction
The QRC API, powered by Nabr, catches Tajweed errors at the phoneme level on live audio: feedback as the user recites, not afterwards.
A research agent that does the research
Tbyaan synthesises the relevant evidence, builds an outline and drafts from classical sources, searching the Quran, authenticated Hadith, Tafseer and Islamic Q&A together rather than four times over.
The numbers behind it
- 0.83Average on Islamic retrieval: ahead of Gemini, Voyage and Cohere
- 0.59%Character error rate for Nabr, live in the Eqraa Quran app
- 354,000+Recitation audio files across 34 reciters
What changes
Infrastructure stops being the project
Fragmented datasets, inconsistent formatting and outdated APIs are somebody else's problem, so the work goes into the part of the app that is actually yours.
Research time goes to analysis
Cross-referencing classical sources by hand is the part that gets removed. The scholarship is not.
Recitation feedback without training a model
A Tajweed-aware correction feature ships without building an acoustic model from scratch first.
How it deploys
Qurani.ai API
The developer interface for Quranic data, semantic search and recitation correction.
MCP server
Connect directly to AI coding environments and describe what you want to build in plain language.
On-premises
The underlying models deploy inside your own environment where an institution requires it.
By Industry
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.