IlmVerse: sourced Islamic answers from scholars, not guesswork

Why Muzamil starts with the problem, not the product

Muzamil Hasan opens by framing a series on AI around two questions: who is building in AI, and what they are building it for. “When building has become easy and comparatively cost-effective, it all boils down to ideas,” he says. “It all boils down to outcomes.” For Muzamil, the test of any product is whether it solves a real problem for someone. A product that does not make anyone’s life easier is hard to monetize later.

He explains why religion interests him as an AI subject. He usually avoids the topic on the platform but studies it privately, and he has begun using large language models for religious research in the same way he once used Google: searching, reading, cross-checking. He also says Claude’s responses on religious questions have become more opinionated and more distinct in personality, and he worries that people in the East may have no alternative to models built elsewhere, carrying the narratives of the countries that built them.

A correction, and what SISL is

When Shahbaaz Khan joins, he makes one clarification right away: he is not the founder of IlmVerse but its project head, and IlmVerse is a global initiative of the School of Islamic Scholars and Leaders (SISL) at Al-Kawthar University. He describes the school as recently established on University Road in Karachi, offering an eight-year program that trains scholars. Its instructors include people with PhDs from Madinah, and the curriculum goes beyond classical subjects into language, physical fitness, and entrepreneurship.

Khan’s view is that a graduate of such a program should not be boxed into the narrow role of “alim.” “You can do a lot of things, and there is a need to do a lot of things,” he says. Muzamil says that a modern academic format for Islamic scholarship, including a bachelor’s program with IT, is worth a conversation of its own.

What IlmVerse is, in Shahbaaz Khan’s words

Khan calls IlmVerse “a trusted AI companion for Islamic knowledge.” A user asks a question and gets an answer with sources, drawn from a curated collection weighted by scholars rather than generic web blogs. His example: someone asking how to pray on the moon. The old books do not address it, but people are working on that question, and the system can retrieve their work.

The project started three years ago. Khan says the team brought together scholars from different backgrounds and deliberately sought out differences of opinion to avoid bias. The team also concluded that training a model in Pakistan was not realistic at the time, given the compute required. So they began with a retrieval-augmented generation (RAG) approach: gather the content, put it in one place, and build a chatbot that retrieves from it.

Eight and a half million pages, and the cleaning problem

Khan says the team now works with about 8.5 million pages of Islamic and authenticated content from more than 3,250 authors, a far larger body than they expected. The retrieval setup combines keyword search, semantic search, and a graph layer, so the system can follow links between a Quran verse, its translations, and its commentaries. Khan argues that clean, structured data of this kind is not available as open source, and that the data layer is the real asset. “Before you want to vaccinate, you first need to know where the data is coming from,” he says.

Muzamil asks about funding, and Khan is candid that sustainability is still open. He describes funding from the university’s research fund, a target of 500,000 users before subscriptions, and a plan in which a free tier stays available while enterprise and institutional use cases help cover costs. “I just don’t want that ten years later we find out that they are coming to collect donations,” he says, explaining why he wants the business model settled early.

Digitization and the work that never shows up in a demo

A large part of the conversation covers what Khan calls the last-mile problem. Many Islamic scholarly books have not been digitized, and Arabic optical character recognition cannot be fully automated. Khan says the team tried several tools, and even Gemini, which he considers the strongest model for reading such material, but a single 400-page book still needed human review on every page. Quranic recitation (qiraat) adds another layer, because different recitation traditions must be recorded and reasoned about separately.

Khan also says the team has started a psychology stream, separating a Muslim perspective on psychology from modern psychology, after finding a body of underread content. He describes a vulnerable case: a user who asked a general chatbot about same-sex attraction received counseling that did not reflect Islamic values. Khan argues that a product built for a specific community has to answer such questions within that community’s framework.

Pushback, fear, and who gets to interpret

Muzamil raises several critiques. Muslim communities have often been slow to adopt technology, though they later absorbed innovations such as gunpowder, the printing press, and the loudspeaker. Khan agrees the pattern exists and says the current AI debate has its own extremes, including framings of AI as the dajjal. He added a disclaimer addressing that framing directly because people had been writing to him about it.

The sharpest exchange comes when Muzamil raises Pope Leo’s statement that AI is a tool and not conscious, and the effort Anthropic reportedly made to change how its models are described. Muzamil argues that children growing up now will debate the rights and nature of AI in ways today’s adults may not be equipped for. Khan says the foundational understanding has to be built now, so the community can answer those questions in context later.

On access, Khan is firm that ordinary people can use good sources well, and that access should not be restricted to scholars. Muzamil presses on whether an AI system can or should replace the authority of a qualified scholar. Khan’s answer is that AI should be used but should not become an authority over the user. “Don’t make it an authority over you,” he says.

Closing thoughts

Near the end, Khan describes how his five-year-old son has been encouraged to ask questions from an early age, and how the family explains religious ideas to him to build a foundation. Muzamil thanks Khan and says he hopes the platform serves the wider Muslim world and the global south, not only Pakistan. The episode closes by inviting viewers to share whether they have used comparable tools, and how they weigh general-purpose models against a sovereign system built for their own community’s needs.