Islamic AI Research – “Building with ihsān, rigour, and responsibility”
Overview
In 2025, the Research & Development (R&D) function at Greentech Apps Foundation (GTAF) focused on strengthening the organisation’s Islamic AI Research, analytical foundations, learning science capabilities, Islamic knowledge infrastructure, and decision-making maturity, while remaining firmly anchored to its mission: helping Muslims understand Islam.
This report documents R&D activity quarter by quarter, attributing work to individual contributors while emphasising team collaboration. The year reflects a shift from ad-hoc experimentation toward systematic research, automation, and long-term capability building, spanning donor analytics, Quran and Hadith technologies, search systems, learning behaviour analysis, and strategic ideation.
R&D Team (2025)
- Riasat Islam — Head of R&D
- Md. Ashraful Haque (Ashraf) — Data Scientist
- Mushfiqur Rahman Talha (Mushfiq) — Software Engineer (Search & IR)
- Mainul Hasan — R&D Intern (AI & Knowledge Systems)
- Ferdaous — Senior Independent Contributor (Islamic Knowledge Graphs)
Q1 2025 (January – March)
Focus Themes
- Learning science foundations
- Donor behaviour analysis
- Early automation and experimentation
Interconnected Learning & Active Learning Foundations
Contributor: Riasat Islam
Collaborator: Ishrat
During Q1, R&D laid conceptual groundwork for interconnected learning across GTAF apps, exploring how users cognitively move between Quran, Hadith, Seerah, and related resources. Parallel work examined active learning principles, particularly how reflective engagement, scaffolding, and retrieval practice can be embedded within Islamic learning apps without compromising spiritual integrity.
These ideas informed internal discussions and future experimentation, aligning GTAF’s product philosophy with evidence-based learning science rather than passive content consumption.
Recurring Donor Behaviour & Retention Analysis
Contributor: Md. Ashraful Haque (Ashraf)
Collaborators: Riasat Islam, Galib
Ashraf conducted a comprehensive analysis of DonorBox data to understand why many Ramadan-acquired recurring donors failed to continue long-term.
Key work included:
- Exploratory data analysis of 1,500+ donors across 86 countries
- Retention, churn, and donor lifetime value (LTV) analysis
- Failed payment diagnosis and acquisition channel assessment
Outcome:
The resulting report became a critical organisation-wide resource, directly informing fundraising strategy, donor engagement workflows, and future automation priorities.
Quran Verse Emotion Detection (LLM-Based)
Contributor: Md. Ashraful Haque (Ashraf)
An LLM pipeline was developed to tag Quranic verses with emotional categories (e.g., Hope, Fear, Mercy, Warning), enabling deeper contextual engagement.
Highlights:
- Processed over 6,200 verses
- Tested multiple LLM models with robust retry and recovery mechanisms
- Achieved ~95% usable coverage
Status:
Completed and handed over to the content team for further refinement and application.
Automated Failed Payment Monitoring (Ramadan-Critical)
Contributor: Md. Ashraful Haque (Ashraf)
In response to live Ramadan failures in recurring donations, an automated system was built to:
- Monitor failed Stripe payments
- Classify failure reasons
- Notify teams via Slack with actionable summaries
Impact:
The system enabled real-time intervention, improving donor recovery and preventing further revenue leakage during a critical fundraising period.
Q2 2025 (April – June)
Focus Themes
- Infrastructure automation
- Knowledge extraction pipelines
- Search experimentation
DonorBox Data Pipeline Automation
Contributor: Md. Ashraful Haque (Ashraf)
A fully automated pipeline was implemented to sync DonorBox data directly into Google Sheets for dashboarding and analysis.
Outcome:
Manual data handling was eliminated ahead of Dhul Hijjah campaigns, significantly improving reliability and operational efficiency.
Hadith Narration Chain Extraction (Phase 1)
Contributor: Md. Ashraful Haque (Ashraf)
A two-stage pipeline was developed to:
- Extract isnād (narration chains) from Arabic Hadith texts
- Match narrators against a 25k+ rijāl database using fuzzy and AI-assisted disambiguation
Achievement:
Urgent delivery within a 10-day deadline, with high accuracy and robust handling of complex narration structures.
Quran Memoriser App — User Behaviour Analysis (Phase 1)
Contributor: Md. Ashraful Haque (Ashraf)
Analysis of memorisation patterns revealed:
- Strong engagement at word-level (tajwīd & pronunciation focus)
- High engagement in Juz 30 and key Surahs
- Clear “valley effect” in mid-Quran sections
Key insight:
Users excel at problem identification but lack tools for structured progress tracking.
SearchDeen & Search Experiments
Contributor: Mushfiqur Rahman Talha (Mushfiq)
Mushfiq led systematic evaluation of search quality across GTAF apps:
- Studied IR metrics
- Built comparison pipelines
- Evaluated SearchDeen against external benchmarks
- Experimented with fuzzy and lexical techniques
This work laid the foundation for later hybrid and vector search experimentation.
Q3 2025 (July – September)
Focus Themes
- Depth over breadth
- Knowledge integrity
- Behavioural insights
Hadith Narrator Identification — Complete Pipeline (Phase 2 & 3)
Contributor: Md. Ashraful Haque (Ashraf)
This phase completed an end-to-end isnād processing system:
- 3,200+ Hadiths processed
- ~15,000 narrator mappings
- 95%+ extraction accuracy
- Conservative QA framework with manual verification flags
Notable:
Total processing cost was kept extremely low through careful optimisation, demonstrating that rigorous Islamic NLP does not require excessive resources.
Quran Memoriser App — Full Behavioural Analysis (Phase 2)
Contributor: Md. Ashraful Haque (Ashraf)
A deep behavioural study identified:
- Distinct learning personas across global user communities
- Strong word-level precision over verse-level engagement
- Front-loading and milestone-driven memorisation patterns
Outcome:
Findings directly informed future product design decisions and learning scaffolds.
Talks & Knowledge Sharing
Contributor: Riasat Islam
Throughout Q3, Riasat delivered several talks in London and Birmingham, helping position GTAF as both a practitioner and thought leader.
Q4 2025 (October – December)
Focus Themes
- Maturity & scale
- Research consolidation
- Strategic clarity
Donor Analytics Dashboard (Production Deployment)
Contributor: Md. Ashraful Haque (Ashraf)
A full donor analytics system was deployed, supporting:
- Tier-based segmentation
- LTV forecasting
- Churn detection
- Marketing-ready exports
Impact:
Manual reporting time dropped dramatically, and fundraising teams gained actionable, timely insight for decision-making.
DonorBox → MongoDB Migration
Contributor: Md. Ashraful Haque (Ashraf)
Collaboration: Backend Team
The donor data pipeline was modernised to a scalable MongoDB-based architecture with daily automated syncs, removing Google Sheets as a bottleneck and future-proofing the system.
Quran & Hadith Analytics Audits
Contributor: Riasat Islam
Riasat conducted analytics audits across core apps, ensuring:
- Measurement alignment with learning goals
- Clear interpretation of engagement metrics
Published Research
Contributor: Riasat Islam (with collaborators)
In 2025, GTAF research was published in a leading HCI journal:
Islamic Lifestyle Applications: Meeting the Spiritual Needs of Modern Muslims
International Journal of Human–Computer Interaction
Taylor & Francis, 2025
This work formally articulated design principles for Islamic apps grounded in psychology, user experience, and spiritual authenticity.
Comprehensive Hadith Isnad Review Paper
Contributor: Md. Ashraful Haque (Ashraf)
A review paper synthesised five decades of computational isnād research, proposing a novel three-level framework and identifying critical gaps that limit real-world deployment. The paper establishes a clear research agenda for the next decade.
Strategic Ideation Workshop — New Frontier 1.1
Contributors: Cross-functional R&D & Product team
Facilitated by: Md. Ashraful Haque (Ashraf)
An organisation-wide workshop transformed ~70 unconstrained ideas into five prioritised strategic initiatives, complete with implementation plans and success metrics, aligning R&D, product, and mission under a shared roadmap.
Here is a new, self-contained section you can insert into the report (Q4 fits best, but it can also sit under a separate Applied Research & Thought Leadership heading). Tone is academic, Islamic, and non-fluffy.
ADAB: Culturally Aligned AI for Responding to Islamic App Reviews
Contributor: Mushfiq and others in collaboration with IUT
Category: Applied Research · AI Ethics · Islamic UX
In 2025, R&D contributed to the conceptualisation and publication of ADAB, a culturally aligned framework for generating automated responses to user reviews in Islamic applications.
The motivation behind this work was the recognition that existing automated review-response systems, while increasingly sophisticated, largely ignore Islamic etiquette (adab), values, and cultural norms. For Islamic applications, user feedback is not merely transactional; it is often tied to religious practice, sensitivity, and trust. Inappropriate or tone-deaf automated responses risk undermining credibility and user confidence.
The ADAB framework proposes a hybrid AI approach that combines:
- Retrieval-Augmented Generation (RAG) to preserve contextual accuracy,
- Aspect-Based Sentiment Analysis (ABSA) to identify specific concerns within reviews,
- Etiquette-aware prompt engineering to ensure responses reflect Islamic manners such as respect, humility, gratitude, and restraint.
The system was evaluated through direct pairwise comparisons against baseline automated responses. Results showed that ADAB-generated responses were preferred in a significantly higher proportion of cases, indicating that culturally and religiously aligned AI systems can meaningfully improve user perception and engagement in Islamic app contexts.
Beyond its immediate application, ADAB contributes to a broader discussion within R&D around responsible AI in religious domains, emphasising that technical correctness alone is insufficient where values, belief, and trust are central. The framework serves both as a practical system design and as a set of guiding principles for future AI-enabled interactions across GTAF products.
Link:
ADAB – Culturally Aligned AI for Responding to Islamic App Reviews
https://gtaf.org/blog/adab-culturally-aligned-ai-for-responding-to-islamic-app-reviews/
Intern & Capacity Building
Internal Knowledge-Base Chatbot
Contributor: Mainul Hasan (Intern)
Mainul built an internal chatbot powered by GTAF documentation and project data, enabling staff to query organisational knowledge efficiently. Alongside technical delivery, the internship emphasised end-to-end product thinking, user testing, and reflective practice.
Islamic Knowledge Graphs
Contributor: Ferdaous (Independent)
Ferdaous contributed senior-level thinking on Islamic knowledge graph construction, focusing on structured representation of scholars, concepts, and textual relationships—work intended as long-term intellectual infrastructure rather than immediate product features.
Closing Reflection
The work carried out by the R&D team in 2025 was undertaken with a strong sense of amanah—the responsibility of working with knowledge, data, and religious texts that are sacred to Muslims. Every system built, analysis conducted, and experiment run was approached with the intention of serving the Ummah, seeking correctness over speed, and restraint over unnecessary novelty.
We recognise that all benefit and success are ultimately from Allah alone. Any good that has come from this work is by His permission, and any shortcomings are from ourselves. We ask Allah to accept these efforts, to grant sincerity (ikhlāṣ) in our intentions, and to guide us to continue this work in a way that brings benefit without harm.
We humbly ask for your duʿāʾ that Allah grants us steadfastness, wisdom, and protection from error, and allows this work to be a means of ongoing benefit (ṣadaqah jāriyah) for those who use it and for those who contribute to it.
Leave a Reply