AI platform · EdTech · 2025–2026
QuranRI: a grounded AI learning platform
An education institution that had been teaching one-to-one since 2008 needed to reach students it could never staff for. We spent eight months building the platform that does it and, more importantly, the guardrails that make it safe to put in front of a child.



The problem
What was actually hard
The obvious build was a chatbot with a religious-sounding system prompt. That build is also a liability: a general model asked a doctrinal question will answer confidently and be wrong, in a domain where being wrong is not a support ticket but a reputational and ethical failure. The real problem was not making an AI that talks about scripture. It was making one that knows the boundary of what it is allowed to say, and stops.
What we did
How it was built
Retrieval first, generation second
Answers are constructed from an indexed corpus of verified source texts rather than from model memory. The system retrieves, then writes from what it retrieved. If nothing relevant comes back, it says so instead of improvising.
A published refusal boundary
The platform explicitly does not issue legal, medical, psychological, financial or religious rulings, and its terms say so in writing. Deciding what a system must not answer took longer than making it answer well, and it is the single design decision we are most often asked to reproduce for clients.
Assessment before instruction
New learners take a Noorani Qaida placement test (name, email, age, then a graded quiz) so the platform teaches at the right level from the first minute instead of guessing. It doubles as a clean qualification step for the human teaching side.
AI to qualify, humans to finish
The platform handles the scalable part: assessment, practice, repetition, answering the same foundational question at 2am without fatigue. Qualified students then route to one-to-one sessions with VIqra's teachers. The AI does not replace the teacher — it protects the teacher's time for the work only a teacher can do.
Built for children, and audited for it
The product is used by under-18s, so it was built to PIPEDA, GDPR and COPPA expectations from the start: parental review, correction and deletion rights, defined retention, and separate privacy and legal contacts. Retrofitting child-safety compliance after launch is close to impossible.
A full product, not a chat window
Multilingual reading with translation and recitation, geolocated prayer times, an authenticated learner area with saved progress, an incognito mode, light and dark themes, and an in-chat feedback control so every weak answer becomes a training signal.
Outcome
What shipped
Described rather than quantified — we only publish numbers a client has signed off in writing.
- Live in production and publicly testable at quranri.com.
- Every answer is generated from an indexed source corpus, with the source attached.
- A written, enforced refusal boundary: no legal, medical, psychological, financial or religious rulings.
- Placement testing routes each learner to the right level, and qualified learners to a human teacher.
- Built to PIPEDA, GDPR and COPPA expectations, with published privacy and terms and a named data protection contact.
- Designed, built, evaluated and shipped by one team over eight months: architecture, prompting, evaluation, web client, mobile client, brand.
More work
Other projects
PPInstalls
PPInstalls: affiliate marketing platform
A cloud affiliate-marketing platform with tracking, payouts and role-based dashboards.
- Next.js
- Firebase
- Firestore
- Node.js
Utrade Logistics
Utrade Logistics: on-demand delivery app
iOS and Android delivery app with live tracking, payments and a dispatch console.
- React Native
- Redux
- Firebase Auth
- Stripe
TraderMind
TraderMind: market analysis platform
Real-time market data, streaming charts and a configurable analysis dashboard.
- Python
- WebSockets
- React
- MongoDB
Have a project like this?
A 20-minute call, then a written scope with a fixed price and dates. Free, and we sign a mutual NDA first if you want one.
