A live Qur'an app with Qalam, an AI assistant whose answers are grounded in Qur'an and Hadith retrieval.
A Qur'anic study app shipped on Google Play. Its centrepiece is Qalam, an AI assistant that answers questions by retrieving from Qur'an and Hadith sources rather than generating from model memory. The app also carries cinematic Qur'an stories and prayer times. Underneath sits a Python service exposing tajweed colour-coded Arabic text, word-by-word translation, grammatical breakdown, and tafsir, with full-text search over the corpus.
Primary technologies
01 / Context
A Qur'anic study app shipped on Google Play. Its centrepiece is Qalam, an AI assistant that answers questions by retrieving from Qur'an and Hadith sources rather than generating from model memory. The app also carries cinematic Qur'an stories and prayer times. Underneath sits a Python service exposing tajweed colour-coded Arabic text, word-by-word translation, grammatical breakdown, and tafsir, with full-text search over the corpus.
Engineering problem
Religious answers are exactly where a model must not improvise. Every response has to trace back to a real source, which makes this a retrieval problem before it is a generation one.
Approach
Built the assistant as a retrieval-grounded system over structured Qur'an and Hadith data, with a full-text search layer beneath it, and carried the work through product management and QA as well as the AI layer.
02 / Architecture
Layer 01
Tajweed colour-coded Arabic text, word-by-word translation, grammatical breakdown, and tafsir, normalised into one queryable structure.
Layer 02
Full-text search over the corpus, so a question resolves to specific passages before any model is asked to phrase an answer.
Layer 03
The response layer. It answers only from what retrieval returned, which is what keeps religious answers attributable.
03 / Engineering judgment
Religious answers are exactly where a model must not improvise. Treating this as a search problem first made every answer traceable to a source.
The same structured data powers the assistant, the reading view, and search — rather than maintaining separate content pipelines per feature.
04 / Stack
Application
AI & retrieval
05 / Capabilities
Qalam — an AI assistant grounded in Qur'an and Hadith retrieval
Cinematic Qur'an stories
Prayer times
Tajweed colour-coded Arabic with word-by-word translation
06 / Challenges
Challenge 01
Problem
A general-purpose model will confidently produce religious claims it cannot support, which is unacceptable in this domain.
Solution
Constrained the assistant to answer from retrieved passages only, so the failure mode becomes "no answer found" rather than a fabricated one.
07 / Outcome
Live on Google Play with the Qalam assistant, Qur'an stories, and prayer times in production.
Next step
I work on the AI layer — retrieval, reasoning, and the plumbing underneath — and hand it over ready to integrate.